Research Appraisals
Evidence-based critical appraisals of the latest medical research, systematically evaluated using Oxford CEBM methodology.
Showing 119 appraisals
PLoS computational biology
Twelve quick tips for AI-assisted coding in science.
While AI coding tools have demonstrated potential to accelerate software development, their use in scientific computing raises critical questions about code quality and scientific validity. In this paper, we provide twelve practical tips for AI-assisted coding that balance the capabilities of AI with the demands of scientific and methodological rigor. We address how AI can be leveraged strategically throughout the development cycle with four key themes: problem preparation and understanding, managing context and interaction, testing and validation, and code quality assurance and iterative improvement. These principles serve to emphasize maintaining human agency in coding decisions, establishing robust validation procedures, and preserving the domain expertise essential for methodologically sound research. These tips are intended to help researchers harness AI's transformative potential for faster software development while ensuring that their code meets the standards of reliability, reproducibility, and scientific validity that research integrity demands.
29 July 2026
Read appraisal →Journal of robotic surgery
From robot assistance to surgical intelligence: global research trends and emerging frontiers of artificial intelligence-enhanced robotic surgery in urology.
To systematically characterize global publication trends, collaboration patterns, intellectual foundations, research hotspots, and emerging frontiers in artificial intelligence (AI)-enhanced robotic and robot-assisted surgery in urology. Publications were retrieved from the Web of Science Core Collection (WoSCC) using a topic search strategy that included three keyword groups: robot-assisted surgery, urologic diseases or procedures, and AI-related technologies. English-language articles and reviews were included. Meeting abstracts, conference proceedings, editorials, letters, non-English publications, studies unrelated to urologic robotic surgery, and studies without substantive AI or intelligent algorithmic content were excluded. RStudio was used for descriptive statistics and annual publication trend visualization. VOSviewer was used for auxiliary bibliometric network construction and visualization. CiteSpace was used to construct collaboration networks, co-citation networks, keyword co-occurrence maps, cluster maps, timeline and time-zone maps, and citation burst maps. A total of 401 records were initially retrieved, and 253 publications were finally included after screening by document type, language, and topical relevance. These comprised 199 articles (78.66%) and 54 reviews (21.34%). Publications in this field began in 2004 and increased rapidly after 2018, reaching a peak of 53 publications in 2025. Because 2026 was an incomplete retrieval year, only 23 publications were recorded. Italy, the United States, the Netherlands, China, and Japan were the leading contributing countries. The University of Turin, Azienda Ospedaliero-Universitaria San Luigi Gonzaga, IRCCS Fondazione del Piemonte per l'Oncologia, the Netherlands Cancer Institute, and Leiden University Medical Center were the most productive institutions. Keyword clustering showed that the major research hotspots included renal cell carcinoma, augmented reality, image-guided surgery, indocyanine green, machine learning, registration, deep learning, and bladder cancer. AI-enhanced robotic surgery in urology has evolved from early research on registration, navigation, and image-guided surgery toward a surgical intelligence stage characterized by augmented reality, three-dimensional reconstruction, machine learning-based prediction, deep learning-based segmentation, surgical video understanding, and skill assessment. Future studies should prioritize multicenter standardized datasets, sharing of intraoperative video and robotic platform data, model interpretability, prospective validation, and integration into real-world clinical workflows.
27 July 2026
Read appraisal →JMIR medical informatics
Large Language Models for Endodontic Symptom Assessment and Treatment Planning Using Image-Free Clinical Records: Comparative Evaluation Study
BACKGROUND: Accurate assessment of pulpal status is essential for achieving successful endodontic outcomes. However, direct evaluation remains inherently challenging because the pulp is surrounded by calcified tissue, necessitating reliance on clinical and radiographic examinations for diagnostic and prognostic decision-making. These procedures demand substantial clinical expertise and time, and less-experienced clinicians often face challenges that may lead to errors in diagnosis and treatment planning. Recent advancements in large language models (LLMs) offer promising opportunities to enhance clinical reasoning by facilitating the integration of evidence and supporting methodical diagnostic decision-making. OBJECTIVE: This study aimed to evaluate the clinical applicability of LLMs by comparing their text-based clinical screening performance and the clinical validity of their treatment plan responses with those of human evaluators. METHODS: Between January 2011 and December 2022, 100 clinical cases involving primary endodontic disease were randomly selected from the clinical records of outpatients who visited the Department of Conservative Dentistry or Advanced General Dentistry (AGD) at Yonsei University Dental Hospital. Four prompt types, combining 2 variables (language and role), were used as input for 4 LLMs. Both LLMs and human evaluators (AGD specialists, AGD residents, endodontic residents, and senior dental students) assessed the cases using text-based clinical records. Radiographic images were not directly provided. Screening performance was evaluated using a 0-to-2-point concordance scale, and treatment plan validity and relevance were assessed using a 5-point Likert scale. RESULTS: Among the 4 LLMs evaluated, ChatGPT achieved the highest mean concordance score on Korean-doctor prompts (mean 0.98, SD 0.82). However, this score did not reach the partially correct criterion of 1 on the 0 to 2-point scale. Clova X recorded the lowest mean score on English-patient prompts (mean 0.23, SD 0.63). Across both diagnostic categories, AGD specialists demonstrated the highest diagnostic accuracy (pulpal: 0.70; periapical: 0.65), with higher sensitivity but lower specificity than those exhibited by the other groups. ChatGPT also showed favorable performance among the LLMs, with accuracies of 0.65 (95% CI 0.55-0.74) for pulpal disease and 0.57 (95% CI 0.47-0.69) for periapical disease, which were comparable to those of AGD and endodontic residents. CONCLUSIONS: Under image-free clinical record review conditions, ChatGPT 4.0 showed relatively higher and more consistent performance in symptom screening and treatment planning compared to the other LLMs evaluated. However, its highest mean score of 0.98 (SD 0.82) did not reach the partially correct criterion of 1 on the 0 to 2-point scale. Hallucinations generated by LLMs and experience-dependent interpretation biases among human evaluators remain key challenges that require attention. Therefore, continuous clinical supervision and comprehensive user training are necessary for the safe and effective clinical application.
25 July 2026
Read appraisal →JMIR medical informatics
Natural Language Processing Applied to Psychiatric Clinical Notes: Scoping Review
BACKGROUND: Psychiatric clinical notes in electronic health records (EHRs) provide rich longitudinal information that can support clinical decision-making. Using historical medical data can enable earlier identification of mental illness, better characterization of disease trajectories, and more personalized treatment planning. Natural language processing (NLP) transforms these unstructured notes into analyzable representations for research and care. OBJECTIVE: This study aims to systematically summarize NLP methodologies for psychiatric clinical notes, compare major modeling paradigms and application areas, and highlight emerging large language model (LLM) trends, key challenges, and future research directions. METHODS: Following the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) guidelines, a literature search was conducted for articles on NLP methods based on psychiatric clinical notes published from January 2021 to December 2025 in Ovid MEDLINE, Ovid EMBASE, PubMed, Scopus, Web of Science, the ACM Digital Library, and ScienceDirect. This scoping review analyzed NLP methods applied to psychiatric clinical notes, focusing on major trends, identifying suitable features for traditional machine learning (ML)-based models, applications of pretrained language models (PLMs), and key challenges. Approaches were categorized as rule-based, traditional ML, hybrid, deep learning (DL), and LLM-based methods across information extraction and text classification tasks. RESULTS: In total, 101 studies were eligible for inclusion. Rule-based methods (n=36) and hybrid approaches (n=34) remained the most widely used techniques, largely favored for their interpretability in handling nuanced, subjective clinical notes. These were followed by DL (n=15), traditional ML (n=10), and LLM-based approaches (n=6). Traditional ML studies relied heavily on engineered features, which could be grouped into 5 broad categories: domain knowledge features, lexical and statistical features, vector-based semantic features, emotion-related features, and temporal features. PLMs improved performance mainly through domain adaptation and task-specific fine-tuning, enhancing the handling of psychiatric language, medical terminology, and clinical note structure. LLM-based studies, although still limited in number, indicated a growing shift toward generative and reasoning-based applications. CONCLUSIONS: Hybrid NLP approaches remain dominant, combining domain rules with ML for extraction and classification. DL approaches continue to advance, with domain adaptation supporting medical terminology and clinical semantics. LLMs may further automate complex workflows via zero-shot capabilities and reasoning, alongside growing interest in temporal modeling and multimodal integration. Key future needs include improved generalizability across institutions, privacy protection, and careful attention to ethical implications in clinical deployment.
13 July 2026
Read appraisal →Health care management science
An online algorithm for integrated scheduling of pre-treatment and treatment appointments in radiotherapy using deep reinforcement learning
Scheduling patients for radiotherapy within the oncology department is vital for ensuring timely and effective treatment. The patient's therapy consists of a sequence of pre-treatment consultations, i.e. preliminary investigations to define the required treatment, followed by the radiotherapy treatment sessions on linear accelerators. Timely scheduling of pre-treatment consultations is crucial to prevent unnecessary delays in radiotherapy treatment, thereby impacting patient health. Therefore, we suggest integrating both scheduling problems. We employ a Reinforcement Learning-based scheduling technique to solve the multi-appointment scheduling problem in an online fashion. We design a reward function comprising three components to guide the agent towards learning an effective scheduling policy involving multiple appointments. Additionally, we compare our proposed model with benchmark online heuristics. We further provide insight into the reward function design, which steers the agent to learn a feasible policy that leads to high-quality solutions. Experiments show that our algorithm outperforms the benchmark heuristics.
12 July 2026
Read appraisal →PloS one
Self-positioning in space science communication: A corpus-assisted discourse study
This study incorporates text mining into critical discourse analysis to examine how government science agencies in China and the United States position themselves in space science communication on social media. Two specialized corpora have been built by collecting posts from government science agencies on Weibo from China and X (formerly Twitter) from the U.S. With the help of the text mining tool KH Coder, this study gives a corpus-assisted discourse analysis of the particular ways of positioning at different levels of discourse: (1) topics/themes, (2) addressing terms, and (3) those words that co-occur with self-addressing terms in a sentence. The findings reveal significant differences in their preferential ways of positioning. Chinese government science agencies present themselves as state-affiliated yet approachable institutions, blending achievements and operational efficiency with patriotism and collective pride. Their use of diverse addressing terms and co-occurrence patterns portrays them as experienced, supportive guides, balancing national identity with interpersonal closeness. In contrast, U.S. government science agencies emphasize professionalism, focusing on research, space exploration, mission execution, and audience engagement, with minimal reference to state affiliation. Their addressing terms are formal and standardized, with pride centred on mission success and discovery, highlighting expertise and scientific leadership. Their preferential ways of positioning are further explained in their respective contexts in order to present a proper understanding of these differences.
12 July 2026
Read appraisal →Journal of the American Medical Informatics Association : JAMIA
Disparate language and model effects on AI-based translation and recognition of genetic conditions.
INTRODUCTION: Artificial intelligence (AI) is increasingly prevalent. Patients and clinicians may use AI-based tools in many different languages. OBJECTIVE: To investigate AI translation tools for descriptions of genetic conditions and how AI identification of genetic conditions is affected by translations. MATERIALS AND METHODS: We used Neural machine translation (NMT) and large language-model (LLM) translation to translate descriptions of 40 genetic conditions into 191 and 93 languages, respectively. Excluding translations retaining English medical terms verbatim, we respectively focused on 139 and 70 languages. After assessing translations, we assessed the ability of 3 proprietary and 3 open-weight general LLMs to identify conditions in the translations. We analyzed how accuracy was affected by the conditions' prevalence in the literature, and attributes of the languages (the script, language family, and prevalence of the language in training sources). We also investigated adaptive translation for select languages. RESULTS: We found significant differences in condition identification based on the translation method, condition, language, and prediction model. The accuracy of some models was more affected than others by factors like the conditions' literature prevalence, language script, family, and language prevalence. Adaptive translation for select languages did not improve translations or diagnostic accuracy with the 3 tested LLMs. However, further analysis with 1 language showed that this approach was more effective with smaller LLMs. CONCLUSIONS: AI-based translation has variable performance, which can affect the ability of AI models to recognize genetic conditions. These findings should inform safe medical AI use to support consistent performance in different languages.
2 July 2026
Read appraisal →Proceedings of the National Academy of Sciences of the United States of America
Large language models accurately identify decision reasons in verbal reports.
Understanding the reasons behind human choices under risk is a central goal of decision scientists, but traditional methods relying on behavioral data are limited by strict invariance assumptions. We introduce a scalable analytical framework using large language models (LLMs) to analyze verbal reports and identify articulated reasons for choice between monetary lotteries. A validated LLM accurately identified predefined decision reasons in participants' free-text reports, aligning with their actual choices in 95% of trials. Our analysis reveals that the reasons behind people's decisions vary systematically and are driven more by the structure of the choice problem than by individual differences. Crucially, reasons identified from verbal reports yield more parsimonious and informative representations of decision processes compared to those inferred from choices alone; furthermore, problem-specific reason profiles achieve out-of-sample prediction accuracy that is competitive with established computational models. This work demonstrates that verbal reports are a rich data source and our analytical framework can unlock their potential, delivering results that challenge the field's foundational invariance assumptions and pave the way for more context-sensitive and interpretable models of human decision making.
1 July 2026
Read appraisal →Journal of medical Internet research
Trusting Generative AI for Health Advice: Preregistered Survey Experiment
BACKGROUND: Generative artificial intelligence (AI) systems are increasingly used for health information seeking, yet it remains unclear how the public evaluates AI-generated health advice relative to guidance from credentialed clinicians in digital environments. Understanding the conditions under which AI is perceived as credible is critical as these systems become integrated into digital health ecosystems. OBJECTIVE: This study examined how source type (a human nurse in an online portal, a health care-specialized generative artificial intelligence system framed as an "AI nurse," or ChatGPT [OpenAI], a general-purpose chatbot), message characteristics, contextual risk, values framing, and individual differences in medical skepticism and experience with AI shape credibility evaluations of the provided advice and its purported source. METHODS: In a preregistered online experiment, a national sample of US adult participants (N=1502) was randomly assigned to 1 of 3 source conditions and evaluated health advice across 3 scenarios: low risk (dietary advice for cholesterol), high risk (chest pain triage), and a morally sensitive scenario (egg freezing). Advice type (intuitive vs counterintuitive) was manipulated in the risk scenarios, and ideological framing (neutral, conservative-leaning, and liberal-leaning) was manipulated in the morally sensitive scenario. Primary outcomes included participants' perceived credibility of the advice and beliefs about whether the patient should follow it. Source-level perceptions of competence and benevolence were also assessed. Medical skepticism and prior AI experience were examined as moderators. RESULTS: Advice attributed to a human nurse was rated as more credible than advice attributed to either AI source. Message intuitiveness showed effects comparable to, and sometimes larger than, the effects of source: intuitive advice was perceived as more credible than counterintuitive advice, with this difference amplified in high-risk contexts. In the morally sensitive scenario, ideological framing influenced perceived bias but did not interact significantly with source. Medical skepticism moderated source evaluations: higher skepticism was associated with greater perceived competence of the AI nurse and lower perceived competence of the human nurse. CONCLUSIONS: Generative AI is evaluated within existing credibility frameworks rather than dismissed outright as inferior to human expertise. While licensed clinicians retain a credibility advantage, AI-generated advice is generally perceived as competent and legitimate. Importantly, individuals skeptical of traditional medical authority may evaluate AI-based guidance more favorably, suggesting that AI systems may redistribute-rather than uniformly erode-trust in health advice. As AI tools become embedded in patient-facing health platforms, message design and audience characteristics may shape acceptance more strongly than source labeling alone.
24 June 2026
Read appraisal →International journal of medical informatics
From algorithmic innovation to clinical deployment: A systematic review of methodological gaps limiting federated learning in healthcare
OBJECTIVE: Federated learning (FL) is a distributed machine learning paradigm designed to enable model training across decentralized data sources without requiring data centralization. This review critically examines FL as a methodological approach in biomedical informatics, summarizing its conceptual foundations, methodological variants, successes, and limitations, and identifying directions for future research. METHODS: We conducted a systematic methodological review following PRISMA 2020 guidelines. Searches were performed in PubMed, Web of Science, Scopus, IEEE Xplore, and Google Scholar for studies published between January 2017 and March 2025. Articles were included if they proposed, extended, or critically analyzed FL methods for biomedical or life science data. Methods were categorized by federation topology, optimization strategy, data heterogeneity handling, privacy mechanisms, evaluation design, and translational readiness. RESULTS: From 8,412 records, 97 articles met inclusion criteria. Rapid methodological innovation exists in optimization and privacy mechanisms, while support for design-time evaluation and governance remains limited. CONCLUSION: Federated learning represents a significant methodological advance in biomedical informatics but current implementations address only a subset of translational challenges. Future work must integrate study design, evaluation, interpretability, and governance into FL methods.
16 June 2026
Read appraisal →Journal of medical systems
MEL-IA: An Interoperable AI System for Multimodal Skin Lesion Classification in Hospital Settings
UNLABELLED: Skin cancer represents a growing healthcare challenge, motivating the development of AI sytems that can support lesion assessment while integrating efficiently into clinical workflows. This study presents MEL-IA (MobilE skin Lesion dIAgnosis), an interoperable AI system designed for automated skin lesion classification and full integration with hospital information systems. METHODS: The system combines dermatoscopic images and structured clinical metadata using an EfficientNet-B4 - based multimodal model trained on the ISIC 2019 BCN_20000 and MSK subsets. Internal robustness was assessed through stratified 5-fold cross-validation, and external generalization was evaluated on the independent HAM10000 dataset. Interoperability and operational performance were validated through deployment in a real hospital environment using HL7, DICOM, PACS, and HIS/RIS systems. RESULTS: We evaluated internal robustness using stratified five‑fold cross‑validation, with global accuracy of 0.86, macro F1-score of 0.85, and AUC values above 0.97. The external evaluation showed that the model generalizes well under different acquisition conditions, achieving an accuracy of 0.84 and balanced accuracy of 0.65, with sensitivities of 0.60 for melanoma 0.72 for basal cell carcinoma. The multimodal model outperformed the image model baseline across all classes. Technical deployment confirmed full interoperability, > 99% successful study integration, and near real time processing. CONCLUSIONS: MEL-IA provides multimodal skin lesion classification capabilities and integrates smoothly into hospital infrastructures based on standards. The results demonstrate technical feasibility, interoperability, and operational viability for deployment within dermatology oriented clinical workflows. Further clinical validation studies are warranted to assess the system's impact on diagnostic decision-making and patient outcomes.
14 June 2026
Read appraisal →Scientific data
Appendix300: Surgical video and patient metadata of 330 laparoscopic appendectomy cases from five institutions
The limited availability of diverse and representative training data poses a critical barrier to the development of clinically relevant computational tools for intraoperative surgical decision support. Surgical procedures are not routinely recorded, and data annotation requires domain expertise, resulting in a scarcity of open-access surgical video datasets with high-quality annotations. Existing datasets are typically limited to single institutions and specific procedures, such as cholecystectomy, and rarely comprise patient-level metadata like demographic characteristics, disease history, or laboratory parameters. The Appendix300 dataset comprises 330 laparoscopic surgery recordings, including 325 full-length laparoscopic appendectomies and 5 control recordings from non-appendectomy procedures in pediatric and adult patients treated at five German centers. The dataset includes patient-level clinical metadata (demographics, medical history, clinical symptoms, preoperative laboratory parameters, and histopathologic findings, as well as standardized expert annotations of the laparoscopic grade of appendicitis). This dataset enables novel validation tasks for computer vision in laparoscopic surgery and facilitates simulation of decentralized learning approaches, overall enhancing the breadth and translational relevance of AI-based surgical video analysis.
14 June 2026
Read appraisal →The pharmacogenomics journal
Development and validation of an open data model for pharmacogenetics to enable semantic interoperability in clinical practice
Pharmacogenetics uses genetic testing to improve the safety and effectiveness of prescribed medicines, yet implementation at scale remains limited due to the absence of interoperable health IT solutions that integrate results into prescribing workflows. This study aimed to develop and validate open data standards for pharmacogenetic results to enable interoperability across healthcare systems. A baseline data model was constructed using the open standard openEHR by synthesising literature, genomic sequencing outputs, and international data specifications, and refined through iterative workshops with the Global Alliance for Genomics and Health. The model underwent two rounds of structured peer review involving 24 experts from 10 countries. Mapping to HL7 FHIR was evaluated using both manual and automated approaches, including the FHIR-Connect tool. The resulting standardised pharmacogenetic data model separates test results from therapeutic implications and incorporates recognised terminologies such as SNOMED CT and HGNC. It achieved international consensus and is published on the openEHR Clinical Knowledge Manager platform. Mapping to HL7 FHIR demonstrated bidirectional information flow within healthcare systems, with automated mapping enabling scalable and reusable transformations. This work provides a framework for storing and exchanging pharmacogenetic test results, supporting semantic harmonisation, interoperability, and integration with clinical decision support systems. Open data standards for pharmacogenetic test results therefore offer a foundation for scalable implementation of pharmacogenetics in routine clinical practice.
14 June 2026
Read appraisal →JMIR formative research
Automated Machine Learning Frameworks for Radiomics: Comparative Evaluation Study
BACKGROUND: Automated machine learning (AutoML) frameworks can lower technical barriers for predictive and prognostic model development in radiomics by enabling researchers without programming expertise to build models. However, their effectiveness in addressing radiomics-specific challenges remains unclear. OBJECTIVE: This study aimed to evaluate the performance, efficiency, and accessibility of general-purpose and radiomics-specific AutoML frameworks on diverse radiomics classification tasks, thereby guiding researchers and highlighting development needs for radiomics. METHODS: A total of 10 public and private radiomics datasets with varied imaging modalities (computed tomography and magnetic resonance imaging), sizes, anatomies, and end points were used. Six general-purpose and 5 radiomics-specific frameworks were tested with predefined parameters using standardized cross-validation. Evaluation metrics included area under the receiver operating characteristic curve, runtime, and qualitative aspects related to software status, accessibility, and interpretability. RESULTS: Simplatab, a radiomics-specific tool with a no-code interface, achieved the best overall balance between performance and computational efficiency, recording the highest average test area under the receiver operating characteristic curve (mean 78.46%, SD 12.22%) with a moderate runtime (1.1 h). However, its performance was not statistically superior to the most intensive general-purpose solutions. Most radiomics-specific frameworks were excluded from the performance analysis due to obsolescence, extensive programming requirements, or computational inefficiency. Conversely, general-purpose frameworks demonstrated higher accessibility and ease of implementation. CONCLUSIONS: While no single framework demonstrated absolute predictive superiority, Simplatab provides an effective balance of performance, efficiency, and accessibility for radiomics classification problems. However, continued efforts are needed to further mature AutoML solutions in the radiomics domain.
14 June 2026
Read appraisal →Journal of medical Internet research
Patient Cognitive Bias in Large Language Model-Supported Health Consultations: Simulation-Based Comparative Study
BACKGROUND: Large language models (LLMs) are increasingly used by patients for health information and preliminary medical advice. In patient-facing consultations, users may present explicitly stated diagnostic preferences or symptom narratives emphasizing a preferred explanation. Such cognitively biased input constrains the diagnostic context available to the model and may systematically steer its reasoning during interactive LLM-supported health consultations. OBJECTIVE: This study aimed to quantify the impact of patient cognitive bias on LLM diagnostic performance in multiturn consultations, assess the effectiveness of prompt-based mitigation strategies and decoding temperature adjustment, and evaluate a dual-system framework for improving robustness under biased interaction. METHODS: We developed a simulated patient agent to generate both unbiased and cognitively biased consultations using 1273 medical question answering dataset United States Medical Licensing Examination cases. Six widely used LLMs of varying capacities were evaluated through 3-round, multiturn dialogues, after which each model produced a final diagnostic judgment based on the complete consultation record. Diagnostic accuracy was the primary outcome. Secondary outcomes included bias-induced accuracy decline (absolute reduction in accuracy under biased vs standard consultations) and bias-influenced error proportion (proportion of incorrect responses aligned with the patient's preferred but incorrect diagnosis). Three prompt-based mitigation strategies and 4 decoding temperature settings were tested. In addition, a dual-system framework was evaluated, in which a conversational foundation LLM conducted patient interaction and history taking (System 1), while a reasoning-oriented LLM (o1-mini) generated the final diagnostic judgment (System 2). In the foundation-only condition, the same LLM performed both interaction and diagnosis. RESULTS: Across all 6 evaluated models, cognitively biased consultations led to marked diagnostic accuracy declines of approximately 7 to 39 percentage points compared with standard multiturn consultations, whereas static single-response tests and standard consultations showed comparable accuracy. Larger deteriorations were observed in lower-capacity models, with some approaching random-guess performance under bias. Errors were frequently aligned with patient bias, with bias-influenced error proportion exceeding one-third across models, indicating systematic conformity rather than random error. Prompt-based mitigation strategies and decoding temperature reduction yielded limited and inconsistent improvements and did not reliably prevent bias-induced performance loss. By contrast, the dual-system framework substantially improved diagnostic accuracy under biased conditions, producing gains of approximately 10 to 39 percentage points across most models and recovering a large proportion of the performance lost due to bias, particularly in lower-capacity systems. CONCLUSIONS: Patient-driven cognitive bias represents an underrecognized behavioral risk in LLM-supported health consultations. Common mitigation approaches, such as prompt engineering or decoding parameter adjustment, provide limited resilience. Explicitly separating conversational interaction from deliberative diagnostic reasoning through a dual-system framework enables more robust diagnostic performance under biased input while potentially preserving patient-facing dialogue fluency by retaining the foundation LLM as the conversational component, offering a scalable design strategy for safer medical AI systems.
14 June 2026
Read appraisal →JMIR research protocols
Effectiveness of Guideline-Based Clinical Decision Support Systems: Protocol for a Systematic Review
BACKGROUND: Clinical guidelines (CGs) standardize care through evidence-based recommendations, while clinical decision support systems (CDSS) can assist in applying these guidelines to individual patients. The scientific basis for the decisions offered by decision support systems is often not explicitly stated or not clearly specified in the literature on CDSS. Therefore, a systematic examination of the literature is needed to map the current state of CDSS, with a particular focus on the integration of CGs. OBJECTIVE: This study aims to systematically collect, describe, and synthesize evidence of randomized controlled studies of interventions using CDSS with a well-defined integration of evidence-based CGs and evaluating direct medical outcomes. METHODS: This systematic review adheres to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) and PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) checklists. The eligibility criteria for this review are defined using the patient, intervention, control, outcome, and study design framework, including studies involving patients with any medical condition or disease. Study interventions need to include guideline-based CDSS, encompassing all types of interventions used for treatment. Each intervention must provide a sufficiently accessible technical description, including the types of data and algorithms used for decision-support procedures. The guidelines used within these CDSS interventions must be derived from a clearly defined, evidence-based guideline development process published by a discernible guideline-producing body. Studies must use a randomized controlled study design. Only studies evaluating the effectiveness of the interventions on direct medical outcomes are included. Web of Science, including MEDLINE, and Scopus will be searched with search expressions aligned with the eligibility criteria. RESULTS: On August 11, 2022, the initial search was conducted on Web of Science and Scopus. From a total of 6203 records, 1347 were removed prior to screening as duplicates, 2506 records were excluded during the first screening step, and 2291 were excluded during the second step. Next, 41 papers were excluded based on full-text review, and 18 papers were finally included in the review following this initial search. This review explores whether CDSS based on CGs can improve clinical outcomes, although their effectiveness may vary depending on various factors. Potential limitations, such as high study heterogeneity, have already been identified. An update of the review has been started in April 2025. CONCLUSIONS: To our knowledge, this is the first rigorous systematic review on the effectiveness of guideline-based decision support systems in which the technical integration and algorithmic embedding of CGs have been described or can be inferred from secondary literature. With this review, we aim to address this gap by providing a detailed analysis of existing research and identifying best practices, challenges, and areas for future investigation.
14 June 2026
Read appraisal →Critical care (London, England)
Large language models for optimizing clinical trial recruitment in ICUs: application to ventilator-induced diaphragm dysfunction
BACKGROUND: Ventilator-induced diaphragm dysfunction (VIDD) is a frequent and under-recognized consequence of prolonged mechanical ventilation in intensive-care unit (ICU) patients. Identifying eligible candidates for clinical trials targeting VIDD remains a major operational challenge. This study evaluates the use of large language models (LLMs) to automate patient prescreening from ICU discharge summaries and estimate recruitment capacity for a future phase 2 trial. METHODS: We developed an LLM-based prescreening pipeline to assess trial eligibility criteria from ICU discharge summaries, which was deployed to screen all 2024 ICU stays. Stays that were flagged as potentially eligible underwent expert adjudication. An enriched set of 50 ICU stays was independently annotated by six clinicians to define a reference standard, which was used to evaluate criterion-level model performances using F1-scores. RESULTS: The best-performing model was GPT-OSS:120B with a criterion-level F1-score of 0.82. When applied to consecutive 1,342 ICU stays from Montpellier University Hospital in 2024, the selected model identified 532 patients with ≥ 3 days of mechanical ventilation. After applying exclusion criteria, 185 patients remained potentially eligible. Expert review confirmed 133 patients as eligible, resulting in a positive predictive value of 72% (95% CI 65-78). The LLM-assisted workflow resulted in an estimated 86% reduction in clinician review time. The LLM-based prescreening pipelines achieved criterion-level F1 scores ranging from 0.73 to 0.82, with GPT-OSS:120B demonstrating the highest performance. CONCLUSIONS: LLM-based prescreening offers a promising approach for identifying trial candidates in critical care, prioritizing candidates for clinician review. Future deployments should include targeted expert validation and ongoing monitoring to ensure safety and generalizability.
13 June 2026
Read appraisal →Med (New York, N.Y.)
Rethinking healthcare data interoperability in the age of large language models
Electronic health records contain extensive real-world clinical data, but their effective use is hindered by data heterogeneity and interoperability challenges. Traditional post hoc standardization is costly and labor-intensive and reduces data granularity. Large language models enable analysis of unstructured data without full harmonization but lack precision for some tasks. We propose a hybrid strategy that combines large-language-model-based analysis of legacy data with prospectively standardized data, offering a scalable alternative that challenges the need for retrospective data harmonization and improves interoperability.
13 June 2026
Read appraisal →Journal of neural engineering
An online brain-computer interface for detecting incongruity in augmented reality applications
Objective.Augmented reality can provide digital information about physical entities presented within its real-world context. However, this information might disagree with the user's expectations due to factual errors in the data or cognitive biases. Such incongruity can impair user experience and undermine trust in the AR system. To address this issue, we propose detecting inconsistencies between physical objects and digital information through hybrid brain-computer interfaces.Approach.We conducted two complementary experiments. First, we implemented a strategy that integrates eye-tracking and brain signals for incongruity detection in an offline study. Subsequently, we assessed our approach in an online study in which participants received immediate feedback on the classification.Main results.The grand average event-related potentials revealed consistent electroencephalographic responses to incongruent augmentations, specifically a centroparietal N400 effect, across both experiments. We could further distinguish between congruent and incongruent information with an average balanced accuracy of 70% in the online study.Significance.These findings demonstrate the feasibility of detecting incongruity online, allowing for autonomous system adaptation, like presenting information in a more accessible format or providing contextual support.
31 May 2026
Read appraisal →PloS one
Three-dimensional shape cues affect human and artificial recognition systems differently
Humans and neural networks use shape and texture information differently. While humans weigh shape heavily in their ultimate classification decision, neural networks are more biased towards texture cues. Many tests of shape vs. texture bias have focused on shape recognition from an object's external contour. However, shape information is also conveyed through internal contours, shading, and attached shadows, especially when an object is viewed from noncanonical perspectives. Using models from ShapeNet, we created datasets of 120,000 texture-substituted images of objects from many viewpoints with and without shading and attached shadows. We tested humans' and several neural networks' ability to classify these objects by both their shape and their texture. Humans were much better at classifying texture-substituted objects by their shape than any network, although these differences were greater when shape was defined only by the external contour than when 3D cues were included. Our findings suggest that networks' texture bias is reduced when 3D cues are included in images. We next tested whether the inclusion of 3D cues benefitted humans and neural networks more for images of objects viewed from canonical or noncanonical perspectives. Consistent with earlier research, we found that 3D cues primarily benefitted humans for noncanonical images. For neural networks, the greatest performance gains were for canonical images. These findings suggest fundamental differences in how humans and networks use shading and attached shadows for object recognition. We argue that humans use these cues to infer objects' 3D structures while neural networks use them as another surface-level cue like texture.
31 May 2026
Read appraisal →JMIR medical education
Performance of Large Language Models on the Brazilian National Medical Education Examination: Comparative Benchmark Study
BACKGROUND: Large language models (LLMs) are rapidly incorporated into medical education and examination preparation; yet, most benchmarking evidence is derived from English-language material. Whether frontier commercial models and Brazilian Portuguese domain-specialized systems perform equivalently on high-stakes Brazilian medical examinations remains unclear. OBJECTIVE: This study aims to quantify and compare the performance of 9 frontier commercial LLMs and 1 Brazilian Portuguese domain-specialized system (Charcot, Voa Health) on 2026 Brazilian National Medical Education Examination (Exame Nacional de Avaliação da Formação Médica [ENAMED] 2026) and to describe the patterns of systematic between-model error as complementary quality signal. METHODS: All 100 items of ENAMED 2026 (99 valid after annulment) were administered to 10 frontier-panel models across 5 independent runs under identical Portuguese prompts (temperature=0; top-p=.95). Commercial models were accessed through a unified OpenRouter client layer (DeepSeek provider-pinned). The primary outcome was mean accuracy against the preliminary key; the secondary outcomes were convergence error (CE), normalized mean response time (NMRT), and intermodel agreement. Accuracy was analyzed with Shapiro-Wilk, Levene, Kruskal-Wallis (ε2), Dunn-Holm post hoc, and a binomial generalized linear mixed model with question and run random intercepts. NMRT excluded Charcot (different stack) and Grok 4 (latency outlier). A total of 7 open-weight and small language models were assessed as a small language model (SLM) substudy. RESULTS: Frontier-panel accuracy ranged from 73.74% (365/495) for GPT-4o-mini to 96.97% (480/495) for Charcot. Accuracy was nonnormal (Shapiro-Wilk, W=0.82; P<.001) with homogeneous variance (Levene P=.26). Kruskal-Wallis showed large between-model differences (H9=47.65; P<.001; ε2=0.97). Dunn-Holm flagged 8 of 45 pairs: Charcot was separable from GPT-4o-mini, DeepSeek v3.2-exp, and Grok 4, but not from the top frontier cluster. The generalized linear mixed model preserved the ranking (all comparators odds ratio<1 vs Charcot; upper CI<1 except GPT-5). A total of 9 items met the default CE criterion; item 77 showed 10-of-10 convergence, later confirmed by Instituto Nacional de Estudos e Pesquisas Educacionais Anísio Teixeira rectification, and sensitivity analysis preserved the CE set (5-20 items). Intermodel agreement was high (Fleiss κ=0.852; Krippendorff α=.852). Among 8 retained commercial models, NMRT correlated positively with accuracy (Spearman ρ=0.74; P=.04); these results are interpreted as descriptive of an orchestration-level latency-accuracy law. SLM accuracy ranged from 47.47% (Gemma 3 4B) to 82.22% (GPT-OSS 120B), with lower agreement (Fleiss κ=0.508) and 17 CE items. The best SLM-panel model lagged every frontier-panel model except GPT-4o-mini, with a 12-15 percentage-point gap against the top cluster. CONCLUSIONS: On ENAMED 2026, a Brazilian Portuguese domain-specialized system ranked first, indistinguishable from frontier commercial cluster and above subfrontier and open-weight systems. Charcot's architecture is not publicly disclosed; these findings should be interpreted as comparative black-box evidence of performance and not as mechanistic evidence of specialization. CE was stable, and it prospectively flagged 1 rectified item, supporting its use as a quality assurance screen.
31 May 2026
Read appraisal →Medicine
A bibliometric analysis of artificial intelligence in anatomy education: Current situation, hot spots, and global trends
INTRODUCTION: Artificial intelligence (AI) is increasingly being integrated into anatomy education, yet a comprehensive overview of the field's development and emerging trends remains lacking. This study identifies research hotspots and trends in AI applications in anatomy education through a bibliometric analysis. METHODS: English-language publications on AI in anatomy education were systematically collected from the Web of Science database. Using bibliometric tools, including R, VOSviewer, CiteSpace, and Excel, data were synthesized to explore global research trends and focal areas. RESULTS: A total of 184 papers published between 2005 and 2024 were included. The United States emerged as the leading contributor in terms of publication output and citation impact. Over the past 2 decades, research focus has shifted from general medical education applications toward AI-driven image segmentation, deep learning, and surgical guidance. More recently, large language models (e.g., ChatGPT) have begun to attract attention in anatomy education. CONCLUSION: This bibliometric analysis maps the intellectual landscape of AI in anatomy education, identifying key thematic shifts and emerging frontiers. The findings provide a foundation for guiding future research directions in this rapidly evolving field.
31 May 2026
Read appraisal →Biomedical physics & engineering express
RADIANT: A fully configurable radiotherapy dose prediction framework
Objective.Treatment planning is a multi-disciplinary effort that requires medical decision-making, specialized training, and access to specialized software. Recently, deep learning has arisen as a powerful method for predicting patient-specific dose distributions. This work presents the radiotherapy dose inference and analysis toolkit (RADIANT), an open-source, fully configurable framework for 3D radiotherapy dose prediction. Built upon the Medical Imaging Segmentation Toolkit, RADIANT supports diverse network architectures, loss functions, and training strategies.Approach.We demonstrate its capabilities on cervical and prostate treatment plans generated with the radiation planning assistant, and on head and neck cancer plans from the American Association of Physicists in Medicine OpenKBP challenge data. For cervical cancer, we trained and compared nnU-Net, FMG-Net, W-Net, ddU-Net, and Swin UNETR using clinical metrics such as dose score, homogeneity index, and percent errors inD95,D98, andD99. Benchmarking was also performed on the OpenKBP dataset using dose score and dose-volume histogram (DVH) score for comparison with top challenge submissions.Results.Our results indicate that RADIANT provides a scalable platform for the rapid development and benchmarking of dose prediction models for selected cancer sites (cervix, prostate, head and neck). For cervical cancer, the best configuration (nnU-Net architecture, mean absolute error loss, cosine learning rate scheduler) achieved a dose score of 1.31 and sub-1%errors inD95andD98on the training set, and a dose score of 1.20 on the test set. The same configuration performed best on prostate data with a dose score of 1.96 on a test set. On the head and neck OpenKBP data, RADIANT achieved a dose score of 2.702 and a DVH score of 1.495, competitive with top challenge results.Significance.This work provides an open-source, fully configurable framework for deep learning-based dose prediction. RADIANT facilitates reproducible experimentation across data preprocessing, model configuration, training, and evaluation for multiple cancer sites and anatomical structures.
30 May 2026
Read appraisal →European radiology experimental
Automated identification of MRI series using a hierarchical modular machine-learning pipeline
OBJECTIVE: The volume and diversity of large MR imaging datasets require efficient automated labelling tools for cataloguing MR series, as manual annotation is impractical and costly. However, relying on DICOM header fields alone is unreliable because sequence descriptors are heterogeneous and locally defined, frequently missing or incorrect, and may be altered or removed during anonymisation. MATERIALS AND METHODS: We developed an AI-based modular model to classify MR series. The pipeline comprises five sequential classifiers (Family, Weighting, Fat Suppression, Contrast, and Others) and was trained and tested on 18,181 MRI series from the multicentre PRIMAGE repository. The dataset was split by patient into 80% training/validation and 20% testing; within the training/validation subset, five-fold cross-validation was used. With the exception of contrast classification, all modules used DICOM tag-based machine learning models (CatBoost/Random Forest), while the Contrast classifier incorporated image analysis using a pretrained single-slice ResNet-50. Ethical approval for the study was obtained. RESULTS: Accuracy was 0.994 for Weighting, 0.984 for Family, 0.959 for Fat suppression and 0.958 for Others; the Contrast classifier recorded 0.841. Overall, the end-to-end classification yielded a weighted F1 of 0.849 (CI: 0.837-0.861) and an accuracy of 0.853 (CI: 0.841-0.865). CONCLUSION: The proposed approach provides a reliable and scalable solution for labelling large, heterogeneous MRI datasets across multiple anatomical regions. The pipeline achieved excellent performance for Weighting and Family classification, solid performance for Fat Suppression and 'Others'. However, Contrast classification remains the main limitation and warrants further refinement and/or additional modules. RELEVANCE STATEMENT: Reliable, automated MRI sequence labelling enables faster, reproducible cohort selection and protocol harmonisation in large archives, supporting downstream clinical research and AI tools while reducing manual curation effort and error. KEY POINTS: Multicentre PRIMAGE cohort (18,181 MR series) enabling evaluation in a heterogeneous setting. Machine learning model combining DICOM metadata and image features. High accuracy in series classification: up to 0.994 on key tasks. A scalable pipeline reduces manual annotation workload for radiologists.
30 May 2026
Read appraisal →PloS one
Adaptive and multi-scale feature fusion for Chinese news headline classification
The rapid growth of online news has led to an explosion of short Chinese headlines, which often suffer from sparse features, limited context, and high ambiguity-posing significant challenges for accurate classification. To address these issues, this paper proposes two tailored deep learning models: ERNIE-AAFF-SECNN for large-scale datasets, which enhances semantic representation via adaptive fusion of multi-layer ERNIE features and improves local feature extraction with SE-empowered CNN; and ERNIE-MSSE-DSCNN for small-scale datasets, which integrates multi-scale SE attention, depthwise separable convolutions, and adversarial training to boost robustness under data scarcity. A large number of experiments have shown that both of these models have achieved the most advanced performance. It is worth noting that the accuracy of ERNIE-AAFF-SECNN on the THUCNews and Toutiao datasets is 1.28% and 0.55% higher, respectively, than that of the lightweight SOTA model TinyBERT. The accuracy of ERNIE-MSSE-DSCNN on a 10% training dataset is 2.62% and 3.76% higher than that of the lightweight SOTA model TinyBERT, respectively. It demonstrates outstanding effectiveness under both standard and low-resource Settings. These results demonstrate that targeted architectural enhancements-such as adaptive feature fusion and multi-scale attention with adversarial training-can significantly improve the accuracy and robustness of short-text classification in practical Chinese news applications.
29 May 2026
Read appraisal →JMIR medical informatics
Multimodal Prediction of Renal Tumor Malignancy From Radiology Reports and Structured Electronic Health Records: Retrospective Cohort Study
BACKGROUND: Accurate preoperative prediction of renal tumor malignancy is critical for guiding decisions and reducing overtreatment, as a substantial proportion of renal masses prove benign. Although radiology assessments and structured electronic health record (EHR) data are routinely used, many tumor-specific descriptors remain embedded in free-text radiology reports and are underused due to extraction challenges. OBJECTIVE: This study aimed to develop and evaluate a multimodal pipeline that integrates structured EHR variables with natural language processing features from computed tomography (CT) radiology reports, including large language model (LLM)-extracted abnormality characteristics and transformer-based report embeddings, to improve malignancy prediction. METHODS: We conducted a retrospective cohort study using University of Florida Health Integrated Data Repository Observational Medical Outcomes Partnership-mapped EHR data from December 2011 to August 2024. Adults with renal tumors were included if they had longitudinal diagnostic documentation consistent with a renal mass and at least 1 preoperative renal CT report; final benign or malignant status served as the outcome. Structured features included demographics, comorbidities, medications, vital signs, and laboratory measurements. From the recent preindex CT report, an on-premises LLM isolated kidney-specific findings and extracted abnormality characteristics. Four locally deployed LLMs were evaluated against manual annotations of 500 reports. Kidney-specific text was encoded using pretrained biomedical transformer models, including radiology Bidirectional Encoder Representations from Transformers (BERT) variants. We evaluated unimodal baselines and multimodal early, middle, and late fusion strategies. Model development used 5-fold cross-validation within the 80% training partition; each fold-specific model was evaluated on the same independent 20% held-out test set, with performance reported as mean and SD across the 5 held-out test evaluations. The primary metric was area under the receiver operating characteristic curve (AUC). RESULTS: The final cohort included 967 patients (n=712, 73.6% malignant). In extraction evaluation, Qwen2.5-32B achieved 88.3% overall accuracy with a 100% extraction success rate and was selected for downstream feature generation. Among unimodal models, the structured clinical variable model achieved an AUC of 0.758 (SD 0.012), kidney-specific text with radiology BERT achieved an AUC of 0.746 (SD 0.058), and abnormality characteristics alone achieved an AUC of 0.716 (SD 0.015). Multimodal fusion models achieved higher descriptive performance than unimodal models. Early fusion achieved the highest AUC (mean 0.813, SD 0.008), and F1-score (mean 0.809, SD 0.030), while late fusion achieved an AUC of 0.805 (SD 0.016). Ablation and interpretability analyses suggested complementary predictive information from structured clinical variables and kidney-specific text embeddings. CONCLUSIONS: Integrating unstructured radiology report text with structured EHR variables achieved higher mean predictive performance than unimodal approaches in descriptive comparisons. Multimodal fusion, particularly early fusion incorporating radiology BERT-derived kidney-specific text embeddings, achieved the strongest discrimination, suggesting potential value of natural language processing-enabled multimodal EHR pipelines for informing preoperative risk stratification.
28 May 2026
Read appraisal →Journal of the American College of Cardiology
Opportunistic Cardiovascular Risk Assessment Using Routine Head CT in the Emergency Department
BACKGROUND: Routine noncardiac computed tomography (CT) imaging may contain information about cardiovascular risk. Head computed tomography (CTH) is among the most common imaging studies, conducted annually in millions of patients. Its utility for cardiovascular risk assessment has not been studied. OBJECTIVES: The purpose of this study was to develop and validate deep learning models for predicting incident cardiovascular disease (CVD) and estimating coronary artery calcium (CAC) scores from CTH, and assess performance against clinical risk factors. METHODS: This retrospective cohort study used data from the Stanford Health Care Emergency Department from August 2020 to August 2024. The CVD cohort comprised 27,990 adult patients without known CVD who underwent CTH. The CAC cohort included 2,313 patients who underwent both CTH and coronary CT angiography. Imaging features were extracted from CTH using pretrained deep learning models. Other risk factors were extracted from electronic health records. Outcomes were incident CVD complications (myocardial infarction, stroke, heart failure) and CAC scores (0, 1-10, 11-100, 101-400, >400 AU). Performance was evaluated using the concordance index (C-index) and area under the receiver operating characteristic curve, compared against the baseline model using the variables of the American Heart Association PREVENT (Predicting Risk of cardiovascular disease EVENTs) risk model. RESULTS: Four percent (1,110 of 27,990) of patients (median age 63.0 years, 51.7% female) experienced cardiovascular events. The CTH model achieved a C-index of 0.82 (95% CI: 0.78-0.85) compared with PREVENT (0.75; 95% CI: 0.70-0.79) with difference of 0.07 (95% CI: 0.04-0.10). For CAC estimation (n = 2,313, median age 65.0 years, 53.5% female), the CTH+PREVENT model achieved a C-index of 0.76 (95% CI: 0.72-0.80) and area under the receiver operating characteristic curve of 0.80 (95% CI: 0.73-0.85) for CAC >100. Of the patients, 15.7% were reclassified; higher-risk patients were younger but with higher prevalence of vascular calcifications (30.2% vs 24.8%, P = 0.001) and brain infarcts (20.1% vs 5.8%, P < 0.001). CONCLUSIONS: Routine CTH scans complement traditional risk factors for cardiovascular risk stratification, identifying subclinical disease in younger patients with favorable risk profiles. Clinical integration could improve CVD detection and prevention without additional costs or radiation.
28 May 2026
Read appraisal →JMIR cancer
Machine Learning-Based Survival Prediction Models for Young Patients With Gastric Cancer: Model Development and Validation Study
BACKGROUND: Despite a global decline in the incidence of gastric cancer (GC), the number of cases diagnosed among younger individuals continues to increase. Several studies have been conducted to develop predictive models of mortality in patients with GC. OBJECTIVE: We developed 3- and 5-year survival prediction models for young patients with GC based on machine learning-based survival modeling approaches. METHODS: Data from 813 young patients (≤50 years) diagnosed with GC between 2013 and 2015 were retrieved from the Gastric Cancer Public Staging Database. Among these 813 patients, data from 569 (70%) and 244 (30%) were allocated to the model training and testing datasets, respectively. Random survival forest, gradient boosting survival analysis, extra survival tree, and the Cox proportional hazards model were applied to predict survival outcomes at the 3- and 5-year time horizons. Model performance was assessed and quantified using the concordance index (C-index) metric. For the machine learning prediction models, the mean C-index values and corresponding 95% CIs were estimated across 100 repeated training iterations. RESULTS: For the random survival forest model, the C-index for predicting 3-year mortality was 95.89% (95% CI 95.80%-95.97%), whereas the C-index for predicting 5-year mortality was 91.82% (95% CI 91.68%-91.96%). In the gradient boosting survival analysis model, the C-index for predicting 3-year mortality was 95.32% (95% CI 95.31%-95.33%), and the C-index for predicting 5-year mortality was 89.98% (95% CI 89.95%-90.01%). For the extra survival tree model, the C-index for predicting 3-year mortality was 95.53% (95% CI 95.46%-95.60%), whereas the C-index for predicting 5-year mortality was 94.60% (95% CI 94.50%-94.70%). In addition, the Cox proportional hazards model showed a C-index of 94.15% for predicting 3-year mortality and 82.26% for predicting 5-year mortality. Tumor stage and tumor size were the primary predictive variables used to train the models for mortality prediction at different time points. Other variables exhibited varying levels of predictive contribution across different time points. CONCLUSIONS: These findings may facilitate the identification of high-risk young patients with GC who may benefit from more aggressive treatment strategies by enabling the prediction of mortality risk at different time points.
28 May 2026
Read appraisal →Journal of medical Internet research
Patients' Perspectives on the Implementation of AI in Radiological Diagnostics: Focus Group Study
BACKGROUND: Rapid developments in artificial intelligence (AI) will enable its widespread use in radiological diagnostics in the near future. Patients will then be confronted with findings generated with the help of AI. Understanding patients' perspectives on the use of this technology is one of the key factors for its successful implementation. OBJECTIVE: This qualitative study aimed to gain insight into patients' reasoning about the opportunities and risks of using AI in radiological diagnosis, to identify prerequisites for its acceptance, and to identify aspects that can promote trust in AI diagnoses, especially in scenarios of high personal concern. METHODS: A total of 7 focus groups were conducted with 34 patients (n=15, 44% female participants) aged between 23 and 85 years (mean 49.06, SD 17.08 y), recruited using purposive sampling strategies. Each focus group was audiotaped, transcribed, and analyzed using the method of structured qualitative content analysis. RESULTS: Study findings show that patients are open to the use of AI in radiological diagnostics. The basic prerequisites for this are (1) scientific evidence of safe outcomes that are more accurate and faster than those without AI; (2) recognizable added value in patient care; (3) transparency in the use of AI and disclosure to the patient; (4) comprehensive, binding measures for quality assurance; and (5) the use of AI solely to support the physician. However, the results indicate that further criteria are important for patients to be willing to choose a radiologist who uses AI and to trust AI diagnoses. In situations where they are personally affected, patients fear that physicians will place too much trust in the AI result and that the physician-patient relationship will become dehumanized. Therefore, the physicians' abilities and functions that inspire trust from the patients' perspective must come into play. These include (1) an independent diagnosis by the physician that takes into account not only the clinical context but also the individuality of the patient, (2) a comprehensible explanation of the pros and cons of using AI for patients and clear communication of AI output, and (3) a humane and empathetic physician-patient relationship, which shows that the physician continues to feel responsible for the patient. CONCLUSIONS: The results of the study underscore that a high quality of the entire "sociotechnical" system is an essential prerequisite for patient acceptance of the use of AI in radiological diagnostics and for trust in AI diagnoses. The further development of AI performance must go hand in hand with the creation of framework conditions for its use that meet patients' expectations of the role of the physician and ensure a trust-building physician-patient relationship. The study provides valuable insights into how such integration of AI into radiological practice can be achieved.
26 May 2026
Read appraisal →Scientific reports
From sleep staging to spindle detection: a case study on end-to-end automated sleep analysis
Automation of sleep analysis, including both macrostructural (sleep stages) and microstructural (e.g., sleep spindles) elements, promises to enable large-scale sleep studies and to reduce variance due to inter-rater incongruencies. While individual steps, such as sleep staging and spindle detection, have been studied separately, the feasibility of automating multi-step sleep analysis remains unclear. In this case study, we evaluate whether a fully automated analysis using validated machine learning models for sleep staging (RobustSleepNet) and subsequent spindle detection (SUMOv2) can replicate findings from an expert-based study of bipolar disorder. The automated analysis qualitatively reproduced key findings from the expert-based study, including significant differences in fast spindle densities between bipolar patients and healthy controls, accomplishing in minutes what previously took months to complete manually. While the results of the automated analysis differed quantitatively from the expert-based study, possibly due to biases between expert raters or between raters and the models, the models individually performed at or above inter-rater agreement for both sleep staging and spindle detection. Our results demonstrate that fully automated approaches have the potential to facilitate large-scale sleep research. We are providing public access to the tools used in our automated analysis by sharing our code and introducing SomnoBot, a privacy-preserving sleep analysis platform.
25 May 2026
Read appraisal →Neuroscience
MSCMH-Net: A multi-scale channel-mixing hybrid network for whole-brain segmentation
Whole-brain segmentation constitutes a fundamental task in medical image analysis, providing quantitative assessment of fine-grained brain regions and serving as a cornerstone for both clinical practice and neuroscience research. Despite its importance, the task is inherently challenging given the numerous brain regions, pronounced inter-class heterogeneity, and sophisticated inter-class spatial dependencies. Accurate whole-brain segmentation requires not only precise delineation of local features but also comprehensive modeling of long-range dependencies and global contextual information. To tackle these challenges, we propose the Multi-Scale Channel-Mixing Hybrid Network (MSCMH-Net), a CNN-MLP hybrid framework integrating convolutional and MLP modules at multiple hierarchical levels. The framework leverages the strengths of CNNs to capture local features and spatial structures, while employing MLPs to model long-range dependencies and global contextual information. For integrating global and local information, a channel-mixing module incorporating an exponential moving average (EMA) fusion strategy is employed. A composite dataset of 106 brain MR scans was used, including 36 from MICCAI-2012, 30 from ADNI and 40 from OASIS. Ground truth labels were annotated and double-checked by experts. Comprehensive experiments conducted on the composite dataset validate that MSCMH-Net achieves competitive results relative to existing approaches.
25 May 2026
Read appraisal →Studies in health technology and informatics
Large Language Models for Automating Conformance to Health-Data Standards: The Interoperability Case of HL7 FHIR and OMOP
Interoperability standards enable different healthcare systems to exchange electronic health information seamlessly, and support large-scale analytics needed for current Secondary uses of healthcare data. Transforming diverse clinical inputs into standards-compliant, analyzable data remains labor-intensive and error-prone. Moreover, the learning curve for their proper exploitation can be steep due to extensive documentation, specific technical requirements, and the need for both semantic and technical integration, making them not fully accessible to practitioners in the healthcare domain unless they undergo extensive training. Our study investigated whether large language models can automate conformance while preserving syntactic validity, semantic correctness, and reproducibility, thereby simplifying the adoption of these standards. Using diabetes measurements as a probe domain, we validated extraction fidelity, coding accuracy against controlled vocabularies, and conformance to the domain's main standards. Our prototype emits FHIR Observations and OMOP MEASUREMENT rows.
24 May 2026
Read appraisal →Studies in health technology and informatics
AI-Integrated EEG Decision Support for Neurocritical Care: A Conceptual and Feasibility Framework
The clinical use of electroencephalography (EEG) for neuro-prognostication in neurocritical care remains limited, despite its ability to provide non-invasive insight into brain activity. This study presents a conceptual framework and early-stage prototype of an AI-assisted EEG decision-support system. Using a literature-informed design and system architecture approach, a web-based prototype was developed to support automated EEG analysis and interpretable brain response reporting. Feasibility testing with four users demonstrated reliable system operation and generation of within-patient decision-support indicators. We highlight the system design and feasibility of operationalizing EEG-based analytics into a clinically interpretable decision-support framework for neurocritical care.
24 May 2026
Read appraisal →Studies in health technology and informatics
Logic Models on Health Information Technology: First Insights from a Cross-Disciplinary Scoping Review
BACKGROUND: Health information technology (HIT) interventions are complex, yet theory-based approaches such as logic models or Theory of Change (ToC) remain rarely applied in medical informatics, although they could clarify assumptions, link activities to outcomes, and support design and evaluation. OBJECTIVES: This scoping review examined how logic models and ToC have been used in HIT research across disciplines. METHODS: Following the PRISMA-ScR guidelines, databases were searched up to August 2025. Eligible studies explicitly applied logic models or ToC to HIT. Data were extracted on HIT type, purpose of using logic models, and reported benefits and challenges. RESULTS: Sixty-nine studies met inclusion criteria, most published after 2020. A majority of studies developed logic models for patient-facing HIT such as mHealth or telehealth. Logic models were mainly used to support HIT design or evaluation and were mostly newly developed. Reported benefits of using logic models included conceptual clarity and better stakeholder communication; challenges involved resource demands and limited generalisability. CONCLUSIONS: Logic models and ToC are increasingly used for HIT but remain inconsistently applied and seldom reused. Standardised reporting and repositories could strengthen theory-informed evaluation in medical informatics.
24 May 2026
Read appraisal →Studies in health technology and informatics
Large Language Models for Health Knowledge Modelling in Data Interoperability: A Scoping Review of Methods, Standards, and Applications
Achieving interoperability in healthcare requires semantic alignment of data with standards such as SNOMED CT, LOINC, and FHIR. Manual ontology mapping is slow and inconsistent. Evolving Large Language Models (LLMs) can automate knowledge modelling and data harmonisation. This review explores how LLMs are applied to enhance or automate health knowledge modelling for data interoperability. Following PRISMA-ScR guidance, PubMed and Compendex were searched. Studies applying LLMs to ontology alignment, terminology mapping, schema integration, or knowledge-graph construction were included. From 166 records, 20 studies met the inclusion criteria. Most used GPT-4 or BERT-derived models with retrieval-augmented or embedding-based methods. Reported gains included automation, improved mapping precision, and scalability across standards such as FHIR, OMOP, and UMLS. LLMs show strong potential for automating semantic interoperability in health informatics. Standardised evaluation benchmarks and explainable frameworks are essential for trustworthy adoption.
24 May 2026
Read appraisal →Studies in health technology and informatics
Training Future Health and Information Technologies Professionals on the European Health Data Space.
The European Health Data Space (EHDS) furthers the EU's digital health transformation by enabling secure access to electronic health data by facilitating environments for health data sharing. The topic of EHDS is relatively new and remains underrepresented in most academic curricula in both medical and informatics fields. This study evaluates the impact of a structured educational intervention to improve awareness, understanding, and preparedness among future health and information technology professionals to engage with EHDS-related systems and policies. Conducted at the University of Oulu, the intervention included training sessions, practical examples, and interactive discussions on topics such as the primary and secondary uses of health data, legal frameworks, and technical infrastructure. The survey's results showed significant gains in students' conceptual understanding of EHDS, particularly in distinguishing between primary and secondary uses of health data. We found that students with a European background reported slightly higher familiarity with EHDS-related rights and frameworks. The findings underscore the importance of embedding EHDS education in academic programs to build a digitally competent, informed workforce capable of supporting EHDS implementation. This intervention contributes to the portfolio of learning materials, developed by the SUSA project to integrate digital health policy and governance into multidisciplinary higher education curricula across the EU.
24 May 2026
Read appraisal →Journal of medical Internet research
Human-AI Interaction in Kidney Transplant Decision Support Systems: Qualitative Study of Patient and Support Person Expectations
BACKGROUND: Artificial intelligence (AI) is increasingly applied in medicine, including clinical decision-making. AI-based decision support systems (DSS) can enhance early risk detection and treatment optimization. However, the perspectives of patients and their support persons on AI-assisted DSS in clinical care, particularly regarding shared decision-making (SDM), remain underexplored. OBJECTIVE: This study investigates the expectations, informational needs, and perceptions of patients who underwent kidney transplantation and their support persons regarding AI-assisted DSS and its influence on SDM in posttransplant care. METHODS: In a longitudinal qualitative study, 36 semistructured interviews were conducted with patients who underwent kidney transplantation and their support persons at a German kidney transplant center. Participants were asked about their views on AI's role in follow-up care, its impact on communication, trust, and decision-making, as well as their informational needs regarding AI-assisted DSS. Interviews were transcribed, pseudonymized, and analyzed using framework analysis. RESULTS: Participants recognized AI's potential to support clinicians by identifying risks of transplant loss, rejection, and infection, and by providing data-driven treatment recommendations. However, they emphasized that final decisions should remain with physicians. A majority of participants (n=28, 78%) expressed concern that AI might depersonalize care and diminish physician-patient communication due to a lack of "human touch." Participants demonstrated limited understanding of AI-based DSS functionality and highlighted the need for simple, accessible educational materials (eg, leaflets) explaining AI operations. While most doubted AI could replicate human empathy, some acknowledged that AI might be perceived as more attentive than time-pressured physicians, offering consistent monitoring and support. Participants consistently stressed that AI should augment, not replace, clinical decision-making. CONCLUSIONS: Patients who underwent kidney transplantation and support persons endorse the integration of AI in follow-up care when it enhances clinical decision-making without supplanting the physician's role. Acceptance and trust depend on transparency, accountability, and preserving the "human touch" in care. The development of educational tools to communicate AI functions and limitations is crucial to empower patients and support persons in SDM processes and to ensure AI complements, rather than undermines, patient-centered care.
23 May 2026
Read appraisal →Journal of medical Internet research
Explainable AI in Cancer Imaging: Scoping Review of Methods, Modalities, and Clinical Integration
BACKGROUND: Explainable artificial intelligence (xAI) is increasingly used in medical imaging to enhance transparency, clinical interpretability, and trust in artificial intelligence (AI)-assisted diagnostics, particularly in oncology. Evidence on how explainability is implemented, validated, and reported in cancer imaging remains fragmented. OBJECTIVE: This scoping review aimed to systematically map research applying xAI methods to radiologic cancer imaging, summarize methodological and clinical trends, and identify persistent gaps in validation and integration. METHODS: We conducted a structured search of PubMed and Scopus (final search executed on October 20, 2025), covering studies published from 2017 to December 2024. Eligible peer-reviewed articles using machine learning or deep learning were analyzed with a focus on xAI components. Data from 371 studies were extracted into predefined categories covering cancer type, imaging modality, AI model, xAI method, terminology, validation, code availability, and decision support system integration. RESULTS: Studies focused primarily on breast (112/371, 30.2%), lung (87/371, 23.5%), and brain (56/371, 15.1%) cancers, with prostate, thyroid, and liver cancers also represented. The primary imaging modalities were computed tomography (139/371, 37.5%) and magnetic resonance imaging (104/371, 28%). Deep learning was used in 70.1% (260/371) of studies, classical machine learning in 18.1% (67/371), hybrid pipeline methods for 10% (37/371), and emerging concept-, prototype-, or causal-based approaches accounted for 1.9% (7/371) of studies. Post hoc xAI methods were dominant (305/371, 82.2%), with visualization (163/371, 53.4%), and feature relevance (111/371, 36.4%) as the most common subcategories. Hybrid post hoc or inherent approaches comprised 12.1% (45/371) and intrinsically interpretable methods 5.7% (21/371). Data sources were mostly public (149/371, 40.2%) or mixed (100/371, 26.9%); 22.9% (85/371) used private institutional datasets, and 7.8% (29/371) did not report data sources. Among validated studies, expert or user-based validation was most common (104/193, 53.9%), followed by mixed methods (74/193, 38.3%), while quantitative metrics (10/193, 5.2%) and clinical knowledge-based (8/193, 4.1%) validation remained rare. Only 17.5% (65/371) of studies provided code and 12.1% (45/371) reported decision support system integration, with few achieving actual clinical deployment. CONCLUSIONS: This scoping review maps xAI implementation across multiple cancer imaging modalities, revealing methodological inconsistency and insufficient validation. Most research emphasizes visualization over quantitative interpretability, and few models are reproducible or clinically implemented. These findings provide an evidence base for researchers, clinicians, and regulators to prioritize standardization of xAI reporting, quantitative validation, and user-centered frameworks to advance trustworthy AI in oncology imaging.
22 May 2026
Read appraisal →JMIR medical informatics
The Impact of Electronic Health Records on Family Physicians During Simulated Virtual Encounters: Exploratory Mixed Methods Study
BACKGROUND: This exploratory study investigated the impact of computer use on physician performance during clinical simulations. Standardized patient (SP) scenarios used in family practice certification examinations were adapted to include the use of the electronic health record (EHR). OBJECTIVE: The goal was to compare the impact of EHR use during simulated virtual patient encounters on resident physicians' and staff physicians' patient-centeredness (PC) and overall clinical performance, as well as to measure the cognitive load (CL) imposed by EHR use. METHODS: Sixteen participants each completed 2 video telemedicine simulations with SPs. One simulation case included limited past medical history for the SP in the EHR, while the other did not. Participants were instructed to completely document the encounter using the EHR. Participants' self-perceived CL was measured using the raw National Aeronautics and Space Administration Task Load Index (NASA-TLX). Video recordings were analyzed for participant PC and overall clinical performance. In addition to interacting with the EHR, multiple participants also conducted internet searches. The proportion of time that participants spent interacting with the computer, either using the EHR or searching the internet, was calculated. Inductive qualitative coding of a subset of video recordings (18 of 32 encounters) was performed, with a focus on signs of stress/CL. All videos were assessed for usability problems. RESULTS: Staff physicians (n=6) scored higher on PC compared to resident physicians (n=10) for both cases, though differences were not statistically significant after correction for multiple comparisons (family-wise error rate). Physicians' overall CL, as measured by the raw NASA-TLX, was not significantly correlated with computer use. Exploratory qualitative data analysis found both verbal and nonverbal signs of stress/CL due to computer use while interacting with the SPs. The proportion of time displaying nonverbal signs of stress/CL was calculated for a subset of participants (6 resident physicians and 3 staff physicians). Participant interpretations of instructions to completely document the encounter using the EHR varied widely. It is likely that participants' usual style of documenting, either primarily during or after patient encounters, impacted their use of the EHR while SPs were present. CONCLUSIONS: Use of the computer during video telemedicine appointments may negatively impact physician PC and overall clinical performance. Exploratory qualitative coding identified both verbal and nonverbal signs of stress/CL when participants interacted with the computer and the patient simultaneously. Increased clinical experience helped to mitigate the negative impact of computer use. If the use of the EHR is included in physician certification examinations, clear instructions regarding which tasks must be completed in the EHR during interactions with SPs should be provided.
20 May 2026
Read appraisal →Physiological measurement
Perioperative hypoxaemia early warning based on high-frequency waveform fusion and deep learning
Objective.The accurate prediction of impending intraoperative hypoxaemia events is paramount for patient safety. Current models relying on structural parameters exhibit limitations, including low sampling frequency, weak temporal coherence, and high data attrition rates. Consequently, long-term sequence, high-frequency waveform data are utilised in this study to construct a deep learning-based hypoxaemia prediction model, by which conventional shortcomings are addressed whilst predictive accuracy and robustness are enhanced.Approach.Respiratory, blood oxygenation, and electrocardiogram waveforms were extracted from the VitalDB database. Two processing methods were employed: extraction of waveform-derived variables to construct numerical datasets; and utilisation of the Gramian Angular Summation Field (GASF) and the four-quadrant matrix method to construct image-based datasets. Through multiple dataset construction methods, five datasets were constructed. An LSTM algorithm was employed, using a 5 min learning window and prediction windows ranging from 1 to 13 min, to conduct a comprehensive performance evaluation.Main results.Optimal performance was achieved by all models within a one-minute prediction window. Highly competitive accuracy (0.932), AUC (0.953), and stability (maximum AUC variance 0.067) were demonstrated by the 'Multimodal image model', whilst the poorest performance was exhibited by the 'structured only model'. Moreover, compared to singlemodal models, multimodal models incorporating structured parameters yield only marginal improvements. Consequently, singlemodal data derived solely from raw waveforms are suggested as sufficient for prediction requirements, by which greater application potential is held due to enhanced computational efficiency.Significance.This study systematically demonstrated the significant advantages of high-frequency waveform data. Results indicate that waveform processing methods based on the GASF and the four-quadrant matrix approach can effectively enhance model stability and performance. Furthermore, the singlemodal model relying solely on waveform data, excluding structured parameters, can meet clinical prediction requirements while reducing computational costs. This study provides novel data structures and methodologies for perioperative hypoxaemia prediction.
20 May 2026
Read appraisal →Scientific reports
A human-centric fuzzy decision support system for medical diagnosis using fuzzy cognitive maps
Medical decision support requires models that remain interpretable under uncertainty while still adapting to evolving data and expert knowledge. Fuzzy cognitive maps (FCMs) are attractive in this setting because they encode concept-level relations in a transparent graphical form. However, static expert-defined FCMs are often too rigid, whereas purely data-driven updates may weaken semantic consistency and drift away from clinically meaningful relations. This paper proposes a human-centric adaptive framework for fuzzy cognitive map-based medical decision support. The proposed Human-Centric Fuzzy Decision Support System (HCFDSS) combines data-driven weight adjustment with an explicit expert-correction operator, allowing the causal structure of the map to be refined without discarding domain knowledge. The method is formulated through a two-stage procedure: an inner reasoning loop updates concept activations for fixed weights, while an outer learning loop updates the weight matrix using a regularized objective together with bounded expert intervention. Under standard smoothness and boundedness assumptions, we establish sufficient conditions for stability of the inner reasoning dynamics and convergence of the outer weight-update operator. A numerical illustration shows how expert feedback can revise clinically important edges and alter the final diagnostic activation in an interpretable way. In addition, the paper reports a reproducible empirical evaluation involving synthetic medical scenarios and three public medical benchmark datasets from the UCI Machine Learning Repository, together with sensitivity analysis for expert intervention, preprocessing validation, robustness checks under missingness and imbalance, and expanded comparisons with mainstream machine-learning baselines. The proposed HCFDSS is positioned not as a replacement for high-capacity black-box predictors, but as a transparent and editable decision-support framework in which clinicians can inspect, question, and refine the learned relations. The main contribution of the present work is therefore methodological and theoretical, while the empirical evaluation clarifies the predictive-interpretability trade-off of the proposed framework under reproducible benchmark settings.
20 May 2026
Read appraisal →JMIR medical informatics
From Flow to Feature Using a Proof-of-Concept Spectral-Driven Machine Learning Approach Using Smart Urinary and Drainage Catheter Systems: Algorithm Development and Validation
BACKGROUND: Current urinary and drainage catheter systems collect fluids for visual inspection or manual sampling, offering limited diagnostic value while being labor-intensive and prone to error. Machine learning (ML) has the potential to automate the analysis of these fluids. However, existing methods rely on complex preprocessing steps, which hinder real-time analysis. OBJECTIVE: We aim to develop and evaluate a fully automated, real-time diagnostic approach for smart urinary and drainage catheter systems by leveraging spectral data and ML to differentiate pathological from healthy excreted fluids without the need for manual preprocessing. METHODS: This study proposes a novel, fully automated approach for smart urinary and drainage catheter systems that uses spectra and ML to extract features from excreted fluids, enabling real-time analysis directly. A total of 454 surgical drainage fluid samples (from 181 patients) and 401 urine catheter samples (from 168 patients) were analyzed using smart catheters and drains equipped with compact mini-spectrometer sensors. The collected spectral data were fed into 3 different ML models: a random forest, a partial least squares discriminant analysis regression, and a convolutional neural network (CNN). Each model aimed to extract features and differentiate between pathological and healthy urine and drainage samples based on the various biomarkers available from previously conducted laboratory analyses. RESULTS: All 3 approaches (random forest, partial least squares discriminant analysis regression, and CNN) achieved promising results, demonstrating the potential of the overall approach. In particular, the CNN models trained on the drainage biomarkers hemoglobin and bilirubin achieved the best results. Matthews correlation coefficient scores of 0.83 and 0.81 were obtained for hemoglobin and bilirubin, respectively, when differentiating between pathological and healthy samples using the extracted spectral features. CONCLUSIONS: This work demonstrates the potential of spectral-driven ML for smart urinary and drainage catheter systems. This approach offers a real-time, noninvasive method for analyzing excreted fluids, paving the way for improved diagnostics and personalized patient care. Further research will explore the optimal ML model for this application.
17 May 2026
Read appraisal →JMIR medical informatics
Multimodal Fusion of Echocardiogram Images and Electronic Medical Records for Heart Disease Screening: Retrospective Algorithm Development and Validation Study
BACKGROUND: Echocardiography is a fundamental imaging modality for the diagnosis of heart disease (HD), but its interpretation remains operator-dependent and lacks standardized, data-driven decision support. Although artificial intelligence has improved image-based diagnosis, the added value and interpretability of integrating routinely collected electronic medical records (EMRs) with echocardiogram (ECHO) for large-scale screening remain underexplored. OBJECTIVE: This study aims to develop an explainable artificial intelligence framework that integrates multimodal data, including ECHO images and EMRs, with derived temporal and clinical features, to enhance the screening and interpretability of heart disease diagnosis. METHODS: In this retrospective single-center study, we analyzed 26,936 patients with linked ECHO and EMR records (1470 with clinically confirmed HD and 25,466 without). We constructed cross-sectional EMR features (demographics, comorbidities, and registration history) and extracted view-specific image features with a simplified Inception-v3 backbone. Five modality-specific feature vectors (4 imaging views + EMR) were concatenated and input to an extreme gradient boosting classifier within a patient-level stratified 5-fold cross-validation framework. To mitigate class imbalance during training, non-HD cases in the training folds were randomly downsampled to a 1:3 HD:non-HD ratio; held-out test folds preserved the original prevalence. Post hoc explainability was provided via Grad-CAM heatmaps for images and Shapley additive explanations analysis for EMR features. RESULTS: For the primary binary task (presence vs absence of HD), the multimodal model achieved an AUC of 0.8147 (SD 0.009) on held-out test folds, compared with 0.7785 (SD 0.014) for the image-only baseline, and 0.7343 (SD 0.009) for the EMR-only baseline. At the clinically selected decision threshold, sensitivity, specificity, positive predictive value, and negative predictive value were 84.6% (SD 1.6%), 72.4% (SD 2%), 15% (SD 2.3%), 98.8% (SD 0.3%) for the multimodal model, versus 82.3% (SD 1.9%), 73.5% (SD 2%), 15.2% (SD 2.1%), 98.6% (SD 0.3%) (image-only), and 77.5% (SD 2.1%), 67% (SD 2.3%), 11.9% (SD 1.7%), 98.1% (SD 0.4%) (EMR-only). Grad-CAM visualizations qualitatively highlighted anatomically and physiologically meaningful regions (eg, valvular structures and abnormal flow jets), while Shapley additive explanations analysis identified age, registration years, sex, and hypertension among the top EMR contributors-findings that align with established cardiovascular risk factors. CONCLUSIONS: Integrating echocardiographic images and EMR data in an explainable multimodal framework yields improved and clinically plausible screening performance for heart disease. Future work should focus on external multicenter validation, quantitative assessment of visual explanations against expert annotations, and prospective studies assessing clinical use and workflow integration.
17 May 2026
Read appraisal →Philosophical transactions. Series A, Mathematical, physical, and engineering sciences
Unexpected benefits of self-modelling in neural systems
Self-models have been a topic of interest for decades in human cognition and more recently in machine learning. Yet what benefits do self-models confer? Here we show that when an artificial network learns to predict its internal states as an auxiliary task, the network becomes simpler, more regularized and more parameter efficient. To test the hypothesis of self-regularizing through self-modelling, we used a range of network architectures performing three classification tasks across two modalities. In all cases, adding self-modelling caused a significant reduction in network complexity. The reduction was observed in two ways. First, the distribution of weights was narrower when self-modelling was present. Second, a measure of network complexity, the real log canonical threshold (RLCT), was smaller when self-modelling was present. These results support the hypothesis that self-modelling has a restructuring effect, reducing complexity and increasing parameter efficiency. This self-regularization may help explain some of the benefits of self-models reported in recent machine learning literature, as well as the adaptive value of self-models to biological systems. In particular, these findings may shed light on the possible interaction between the ability to model oneself and the ability to be more easily modelled by others in a social or cooperative context. This article is part of the theme issue 'World models in natural and artificial intelligence'.
17 May 2026
Read appraisal →Physics in medicine and biology
ReaderAdaptNet: modeling reader variability in breast imaging with reader-specific embeddings
Objective.Inter-reader variability remains a major challenge in breast imaging interpretation, particularly for ordinal classification tasks such as breast density and background parenchymal enhancement (BPE). These visual assessments are prone to inconsistencies due to subjective interpretation, limiting the reliability of AI-based models trained on aggregated or noisy labels. This work aims to account for this variability rather than collapse it.Approach.We propose ReaderAdaptNet, a reader-adaptive network that explicitly models inter-reader variability through reader-specific embeddings. A novel two-stage deep learning framework underpins the training process of the model. The first stage jointly learns discriminative image features and low-dimensional embeddings representing individual annotation styles, enabling personalized classification. The second stage allows for embedding calibration using a small set of labeled examples, enabling rapid adaptation to new readers or consensus definitions without retraining the full model. The method is designed for multi-reader datasets and is evaluated on two breast classification tasks: breast density classification from full-field digital mammography and BPE classification from contrast-enhanced mammography.Main results.Results demonstrate that reader embeddings improve both individual and consensus-level performance, and calibrated embeddings enable flexible, low-cost personalization. Reader-specific embeddings of dimension 32 improved the mean classification accuracy across readers, increasing from 76.4% to 84.4% for breast density, and from 65.1% to 72.1% for BPE, compared to the proposed baseline method.Significance.By disentangling stable image features from reader-specific decision tendencies, ReaderAdaptNet provides a parameter-efficient, interpretable route topersonalization(alignment to a given reader) orunification(alignment to an institutional standard) under real-world variability.
16 May 2026
Read appraisal →BMC medical imaging
Development and validation of a novel deep learning coronary artery plaque quantification model
BACKGROUND: Coronary Computed Tomography Angiography (CCTA) is an established tool for assessing coronary artery disease. CCTA determined total coronary artery plaque volume predicts cardiovascular events, but manual quantification is impractical for routine use. Deep learning-based methods offer a promising solution for automated plaque volume assessment. METHODS: We developed and validated a novel software, QuantiPlaque, powered by deep learning models, for automated segmentation of total and calcified coronary plaque volume. Expert manual annotation served as the reference standard. Correlations between deep learning and expert segmentation were evaluated using intraclass correlation coefficient (ICC), Pearson's r, and Spearman's rho, and agreements were evaluated with Bland-Altman analyses. RESULTS: A total of 115 CCTA scans were included. Mean (range) age was 58 (51-64) and 51% were females. The mean coronary artery calcium score was 56 Agatston units, ranging from 0 to 601. The model demonstrated strong correlation and agreement with expert annotation for per-patient total plaque volume (ICC 0.95, mean difference - 8.35 mm3, 95% limit of agreement - 102 to 85 mm3, Spearman's rho 0.87, Pearson's r 0.95) and calcified plaque volume (ICC 0.90, mean difference - 0.28 mm3, 95% limits of agreement - 21.97 to 21.42 mm3, Spearman's rho 0.93, Pearson's r 0.90). Per-vessel analysis showed strong correlation in the left anterior descending artery and right coronary artery but was weaker in the left circumflex artery territory. CONCLUSION: The evaluated deep learning model provides accurate quantification of total and calcified plaque burden, with strong correlation and agreement to expert annotation.
16 May 2026
Read appraisal →Computers in biology and medicine
Dual-model deep learning for Alzheimer's prognostication
Disease-modifying therapies for Alzheimer's disease demand precise timing decisions, yet current predictive models require longitudinal clinical observations and provide no uncertainty quantification-rendering them impractical at the critical first-visit encounter when treatment decisions must be made. We developed PROGRESS (PRognostic Generalization from REsting Static Signatures), a dual-model deep learning framework that transforms a single baseline cerebrospinal fluid (CSF) biomarker assessment into actionable prognostic estimates without requiring prior clinical history. The framework addresses two complementary clinical questions: a probabilistic trajectory network predicts individualized cognitive decline parameters with calibrated uncertainty bounds that achieve near-nominal coverage, enabling honest prognostic communication rather than false precision; and a deep survival model estimates time-to-conversion from mild cognitive impairment to dementia. Using data from over 3000 participants across 43 Alzheimer's Disease Research Centers in the National Alzheimer's Coordinating Center database, PROGRESS substantially outperforms existing approaches including Cox proportional hazards, Random Survival Forests, and gradient boosting methods for survival prediction. Risk stratification identifies patient groups with seven-fold differences in conversion rates, enabling clinically meaningful treatment prioritization. Leave-one-center-out validation demonstrates robust generalizability, with survival discrimination remaining strong across all held-out clinical sites despite heterogeneous measurement conditions spanning four decades of assay technologies. By combining superior survival prediction with trustworthy trajectory uncertainty quantification, PROGRESS bridges the gap between biomarker measurement and personalized clinical decision-making-providing the prognostic timeline that current staging approaches cannot offer.
15 May 2026
Read appraisal →Philosophical transactions. Series A, Mathematical, physical, and engineering sciences
Cognitive glues are shared models of relative scarcities: the economics of collective intelligence
Collective intelligence has emergent problem-solving capacities that are different than those of its subunits. The plethora of multi-scale systems within nature and society makes it imperative to understand the interaction policies necessary and sufficient for emergence of collective intelligences. The economy is a complex system consisting of autonomous elements at multiple scales and which exhibits adaptive problem-solving capabilities, suggesting that the economy offers an interesting, important example of collective intelligence. We identify the price system as the cognitive glue of the economy by acting as a coordinating affordance that enables members to form plans that are mutually compatible. Using the collective intelligence framework of Technological Approach to Mind Everywhere, we elaborate on various aspects of the economy that make it useful to model the economy as a collective intelligence. We argue that any cognitive glue must solve the same kind of problem that the price system solves in broadly the same way that the price system solves it, and thus the price system serves as a generic template or abstract model for all cognitive glues. Finally, we describe some research ideas that combine concepts from biology and economics in the hopes of inspiring interdisciplinary collaboration. This article is part of the theme issue 'World models in natural and artificial intelligence'.
15 May 2026
Read appraisal →The Journal of clinical ethics
Ethical Responsibility in the Off-Label Use of AI in Medical Imaging
AbstractArtificial intelligence in medical imaging (AI-MI), categorized by the U.S. Food and Drug Administration (FDA) as "Software as a Medical Device," can offer significant benefit to medical care. FDA approval-represented by a "label"-indicates that the device has demonstrated safety and efficacy for a specific clinical task and population for which the device is intended. Off-label use of AI-MI outside of the task and/or population can be medically and ethically problematic, yet benefits may be reduced or missed if entirely prohibited. Ethical responsibility frameworks can assist in evaluating and establishing boundaries and opportunities in off-label use of AI-MI. In this work, we review three different models for ethical responsibility and apply them to off-label use of AI-MI. We find that a future-looking collective responsibility framework, involving stakeholders such as industry developers, healthcare organizations, clinicians, and, most importantly, patients and/or their representatives, promotes proactive and practical actions to help ensure safe and ethical use of AI-MI off-label. By adopting this approach, stakeholders can balance innovation with safeguards, advancing patient-centered care while mitigating potential harms associated with off-label AI-MI use. This framework is crucial as AI-MI becomes increasingly integral to clinical decision-making processes.
15 May 2026
Read appraisal →Neuroscience
Recent trends in electrodermal activity signal processing and deep learning methods for emotion recognition
Electrodermal activity (EDA) has emerged as one of the most informative physiological signals for emotion recognition due to its close association with sympathetic nervous system activation. Although research on affective computing has expanded rapidly in recent years, systematic investigations of signal-processing techniques specifically tailored to EDA remain limited. This gap is significant because the selection of decomposition and signal processing strategies directly influences the accuracy, robustness, and interpretability of emotion recognition systems. Furthermore, defining emotion itself and tracing its physiological manifestations through EDA, from raw measurements to recognized affective states, remains an open challenge that highlights the need for methodological rigour. In this review, we synthesise developments published between 2018 and 2025, with a dual focus on advanced EDA signal processing methods and deep learning (DL) approaches for emotion recognition. We review common EDA acquisition sites, discuss challenges arising from non-stationarity and inter-subject variability. The review systematically compares time-domain, frequency-domain, time-frequency, and advanced time-series analysis techniques, alongside emerging end-to-end DL architectures for affective modelling. Unlike prior surveys that emphasise system-level design, this work adopts a signal-processing-centric and physiology-informed perspective. Additionally, we provide a structured comparative evaluation of widely used EDA decomposition methods using key performance metrics, offering a unified framework for method selection and interpretation. Overall, this review serves as a comprehensive resource for researchers and practitioners while advocating hybrid approaches that integrate the interpretability of classical signal processing with the predictive power of predictive methods.
15 May 2026
Read appraisal →PLoS computational biology
Teaching artificial intelligence through drug-drug interaction clustering analysis: Integrating project-based learning and large language models.
In recent years, artificial intelligence (AI) has increasingly influenced daily life and scientific research. Traditionally, AI-related courses have targeted computer science majors, while systematic instructional opportunities for early-stage undergraduates from non-computing backgrounds remain limited. To bridge this gap, we developed an AI course that integrates project-based learning with large language models (LLMs). Specifically, we designed four progressive assignments based on our research project (i.e., drug-drug interaction network clustering analysis). The course does not require prior knowledge of pharmacology or programming. Instead, LLMs are used as assistive tools to support programming, data analysis, and result interpretation. Students engage in a complete workflow, including data curation, algorithm implementation, and critical evaluation of results. Preliminary feedback shows that this approach supports the development of problem-solving skills and increases student engagement. This study provides a framework for integrating LLMs into project-based learning. We believe that this teaching proposal will be valuable and inspiring for educators seeking to design or enrich similar courses.
15 May 2026
Read appraisal →Scientific reports
Federated learning with swarm intelligence for efficient and secure medical image analysis
Collaborative learning in healthcare faces challenges, including strict regulations and fragmented data. This research introduces a federated learning framework that employs swarm intelligence to augment communication and enhance the analysis of medical images. The method optimizes hyperparameters, selects features, and assigns aggregation weights to federated clients simultaneously by combining Particle Swarm Optimization (PSO) and the Firefly Algorithm (FA) with deep Convolutional Neural Networks (CNNs). The framework was tested on three medical datasets: COVID-19 chest X-rays (5,856 images), monkeypox skin images (569 images), and breast cancer mammograms (320 images). These datasets were shared among four fake healthcare institutions. It strives to strike a balance between privacy, communication costs, and classification accuracy. The results showed that the test was 96.71% accurate in detecting COVID-19, 96.06% accurate in classifying monkeypox, and 97.0% accurate in diagnosing breast cancer. The framework was able to handle noise and attacks from individuals who sought to disrupt it, which reduced communication rounds by 25-30%. A privacy-utility analysis revealed that there were acceptable trade-offs, with accuracy remaining above 94%. This study employs robust privacy measures and statistical validation. It also shows how to use medical AI in smaller healthcare settings without putting patients' privacy at risk.
14 May 2026
Read appraisal →Journal of medical systems
Clinical Plausibility in Large Language Model Robustness Testing for Medicine: A Scoping Review
Large language models (LLMs) show promise in medical applications, yet their translation into clinical practice requires rigorous validation. Current robustness testing often employs adversarial approaches borrowed from AI safety, raising questions about their alignment with authentic clinical scenarios. To systematically map methodologies used for robustness testing of LLMs in medical contexts and assess their clinical plausibility. A scoping review was conducted following PRISMA-ScR guidelines, searching PubMed, Embase, Web of Science, IEEE Xplore, ACM Digital Library, arXiv, and MedRxiv from January 2023 to September 2025. Two independent physician reviewers screened 5,331 articles, extracting data on testing methodologies, medical domains, expert involvement, and clinical plausibility. Thirty-three studies met inclusion criteria, predominantly from 2025 (82%). The most common robustness testing approaches were misleading prompts (49%) and adversarial prompts (39%). Only 33% of studies designed tests clearly mimicking plausible clinical scenarios. While 58% reported expert involvement, the depth of integration varied considerably. Studies predominantly addressed mixed medical domains (73%) rather than specialized fields. The emerging literature suggests that LLM robustness testing in medicine often emphasizes technical vulnerability detection, with fewer studies examining clinically plausible scenarios of routine use. Future frameworks should complement adversarial testing with clinically grounded, longitudinal, and specialty focused evaluations to support deployment-relevant inference.
13 May 2026
Read appraisal →Integrative psychological & behavioral science
The Human Mind in Dialogue With its Replica AI - a Historical Perspective
Although artificial intelligence has exploded for the last couple of years, the idea of comparing the human mind with a computer is not new. One may say that Alan Turing started this movement, and this idea got off to a flying start in psychology through the symposium arranged by the "Special Interest Group in Information Theory" at Massachusetts Institute of Technology Sept. 10-11, 1956. This symposium was sponsored by IBM, and Noam Chomsky presented for the first time his ideas about a transformational generative grammar. There were several contributors to the cognitive revolution, but Chomsky's presentation of a linguistic system with a level of precision comparable with mathematics was symptomatic for how a cognitive model for the human mind aimed to be conceptualized. This has brought the question about artificial intelligence to the core of psychology, namely how it can be understood as a model for the human mind. Psychology and its understanding of the human mind have changed along the last 500 years of the Western psychology. After a short presentation of some of the premises for artificial intelligence of today and some aspects of the idea behind the computational mind, these perspectives are compared with the very early use of the term 'psychology' in the early 16th century. The role of classical logic as a model for the human mind was at stake back then and one of the main factors that paved the way for psychology was that Aristotelian logic was criticized and questioned as a model for the human mind. On the other hand, the importance of classical logic has never been completely outdated Consequently, for the last 500 years Western psychology can in many ways be characterized as an incessant battle about the role of logic in the understanding of the human mind. However, the conclusion is that AI of today presupposes an understanding of the human mind that has been outdated for more than 500 years.
13 May 2026
Read appraisal →Scientific reports
Uncovering advanced transfer learning strategies for deep neural networks in natural language processing
In the rapidly evolving landscape of Natural Language Processing (NLP), transfer learning has emerged as a game-changing methodology, fundamentally altering how machine learning models are trained and deployed. The study at hand dives deep into the intricacies of transfer learning techniques, specifically focusing on their application in complex deep learning architectures within the NLP domain. By exploring a variety of architectural designs, fine-tuning methodologies, and alternative training paradigms, we aim to demystify the optimal strategies for harnessing the power of pre-trained models. To quantify the effectiveness of these approaches, we conducted a comprehensive series of experiments targeting key NLP tasks, such as text classification and language generation. Transfer learning significantly accelerated training and improved model accuracy, highlighting its practical advantages in NLP tasks. This dual benefit underscores the immense potential of advanced transfer learning techniques, making them an indispensable tool for future NLP applications. By implementing state-of-the-art transfer learning methodologies, companies can offer faster and more accurate NLP solutions, aligning perfectly with the ethos of "Business Digitalized" and catering to a broad spectrum of client needs in the market.
13 May 2026
Read appraisal →Medical education online
Using randomization to compare AI and expert-generated formative assessment questions in medical education
BACKGROUND: AI-generated content is being used across the education spectrum and is beneficial for creating complex multiple-choice questions, such as those used in medical education. However, evaluating AI-generated content is challenging, and existing testing and evaluation methods are falling short. This study uses randomization to compare medical students' performance on and subjective evaluation of AI versus expert-generated questions. We hypothesized that there would be no difference in student performance or subjective evaluation of AI vs expert-generated questions. METHODS: We designed a single-center, randomized study where medical student participants received either one AI- or expert-generated question per day over four weeks. RESULTS: Our study showed that participants had similar perceptions of AI versus expert-generated questions (p = 0.18), with no significant difference between the distributions of the proportion of correct responses. The cumulative proportion of questions answered correctly over 28 days was consistent across the two question sets. However, participants rated 53% of AI-generated questions as very easy or easy compared with only 31% of the expert-generated questions. DISCUSSION: Randomization was crucial to show that AI-generated questions were nearly imperceptible from expert-generated questions, demonstrating the need for additional evaluation methods to compare AI- and expert-generated medical education content.
12 May 2026
Read appraisal →Physiological measurement
A multimodal fusion network for heart sound abnormality detection and classification
Objective.Accurate physiological assessment of cardiac function from heart sounds remains challenging due to background noise, variable heart rates, and the need for reliable cardiac-cycle segmentation. This study aimed to develop a fully E2E deep learning framework that extracts diagnostic information directly from raw heart sound recordings for cardiac abnormality detection and classification.Approach.We propose HS-MMNet, an E2E multi-modal deep learning framework designed for physiological heart sound analysis. Recordings are preprocessed (normalization and 25-400 Hz bandpass filtering) and divided into fixed-length 2.5 s segments. A Convolution Head with multi-atrous spatial pyramid and channel-spatial attention extracts fine-grained local temporal patterns from the filtered 1-D waveform. A Transformer Head captures long-range spectro-temporal dependencies from Log-Mel spectrograms. These hypotheses are iteratively fused by a novel multi-hypothesis cross-attention module with cyclic query-key-value assignment and a hypothesis-mixing MLP, enabling rich cross-site interaction and effective suppression of noise and non-informative regions. Recording-level classification is obtained via a fully connected layer.Main results.On the PhysioNet/CinC Challenge 2016 dataset, HS-MMNet achieved 94.80% accuracy, 92.10% sensitivity, 96.85% specificity, 87.50% precision, and 89.74%F1-score, outperforming all previously reported methods. On the balanced five-class Yaseen dataset (normal, aortic stenosis, mitral regurgitation, mitral stenosis, mitral valve prolapse), it attained 99.60% macro-averaged precision, recall, andF1-score with only four misclassifications in 1000 recordings, establishing new state-of-the-art (SOTA) benchmarks.Significance.HS-MMNet represents an advance in automated physiological measurement from heart sounds. By eliminating cardiac cycle detection and multi-channel requirements while achieving SOTA diagnostic performance, it provides a practical, scalable solution for accurate cardiovascular screening with primary-care and low-resource settings.
12 May 2026
Read appraisal →JMIR medical informatics
Cough Audio Recognition for Early Detection of Respiratory Diseases: Algorithm Development and Validation Study
BACKGROUND: Coughing is a common clinical symptom and a protective respiratory reflex closely associated with various respiratory system diseases. The acoustic characteristics of cough sounds are influenced by underlying pathological factors, with distinct acoustic signatures corresponding to different etiologies. Through rigorous analysis of these sounds, rapid identification and preliminary diagnosis of related conditions may be achieved. This approach holds great potential for broad application in mobile health and ubiquitous health platforms. OBJECTIVE: This study aimed to explore the application of acoustic analysis of cough sounds in the diagnosis of respiratory diseases to enhance the diagnostic efficiency of health care professionals. METHODS: In this study, we conducted extensive data collection, including voluntary cough audio recordings from patients diagnosed with respiratory diseases (eg, chronic obstructive pulmonary disease, lung cancer, COVID-19, and pneumonia) and from healthy participants. A total of 2610 audio samples were collected. We incorporated a channel attention mechanism (CAM) into the final convolutional block of each residual block in the ResNet18 neural network, thereby constructing the CAM-ResNet18 neural network model. The recorded cough audio samples were converted into spectrograms to form the input dataset for model training. The CAM-ResNet18 model was trained on the training set of this dataset, with iterative parameter adjustments until convergence was achieved. Finally, spectrograms from the test set were fed into the pretrained model for accurate classification of the cough-related conditions. RESULTS: Experimental results on the collected audio dataset demonstrate that the proposed CAM-ResNet18 model achieves an accuracy of 83.9% and an average F1-score of 82.52% in classifying 5 types of cough sounds. In comparison, the traditional ResNet18 model achieves an accuracy of 78.16% and an average F1-score of 78.29%, indicating a clear performance improvement with the integration of the CAM. CONCLUSIONS: The experimental results validate the effectiveness of the proposed method, highlighting its significant potential for application in clinical diagnosis.
12 May 2026
Read appraisal →Neuroscience
In silico neuronal morphology classification: A systematic review
Advances in connectomics and the characterization of neuronal diversity have been fundamental to understanding how the brain works. Defining a taxonomy is still challenging and requires complex computational methods. In this paper, we present a systematic review of the state-of-the-art methods for neuronal morphology classification, shedding light on trends and future directions. We address three core research questions: (1) What computational methods are used to classify neuronal morphologies? (2) What kind of representation do the classification models work on? (3) How are the classification models evaluated? We applied a structured protocol including peer-reviewed studies published since 2018. We collected 840 papers from five databases, published between 2018 and 2024. We assessed risk of bias using predefined quality criteria on study design, methodology, and reporting; 35 studies met these standards and were included to extract information and consolidate the results. The results show that Artificial Neural Networks are currently the main method for classifying neurons based on morphometric features (21 papers). Representations based on microscopy images for extracting morphometric features (30) are commonly used. Accuracy (29) is the predominant evaluation measure. Our research findings contribute to the body of knowledge by providing an up-to-date perspective on neuronal classification, which can guide future research in neuroscience and related fields.
11 May 2026
Read appraisal →PloS one
Reducing annotation burden in medical imaging with ADGNET: A semi-supervised deep learning strategy
We propose ADGNET, a semi-supervised framework for Alzheimer's disease (AD) diagnosis that jointly optimizes image reconstruction and classification through shared feature representations. The architecture integrates a residual backbone with attention modulation for dynamic feature selection, an encoder-decoder reconstruction branch for unsupervised representation learning, and a classification branch with focal loss to address class imbalance. This dual-task design enables effective feature learning from limited annotations. On two public MRI datasets-KACD (2D, 6,400 images) and ROAD (3D, 532 scans)-ADGNET achieves average performance improvements of 4.1% and 7.2% over state-of-the-art methods (ResNeXt WSL, SimCLR) across six metrics. Interpretability analysis using Grad-CAM and attention visualization confirms that the model focuses on clinically relevant neuroanatomical structures, particularly the hippocampus and temporal lobes, with strong correlation to established AD pathology (r = 0.67, p < 0.001). These results validate the model's exceptional generalization capability and feature representation effectiveness across multi-modal medical imaging data, offering an efficient solution for few-shot medical image analysis.
10 May 2026
Read appraisal →Artificial intelligence in medicine
Size doesn't matter: Assessing the trustworthiness of large language models in medical contexts: A focus on epidural information retrieval.
BACKGROUND: Since the release of ChatGPT, numerous LLMs have emerged, providing easy access to information without the need for technical expertise. However, relying on these systems can influence important life decisions, such as the choice to use epidural analgesia during childbirth. Epidural analgesia is widely regarded as the "gold standard" for pain relief during childbirth. However, limited access to anaesthesiologists and gaps in knowledge may prompt individuals to seek information from unverified sources, including AI systems. Misinformation in this area can discourage the use of effective analgesia, highlighting the need to assess the accuracy of LLM-generated content. OBJECTIVE: To evaluate the reliability of LLM-generated information regarding epidural analgesia. METHODS: We posed 10 standardized questions about epidural analgesia to 12 LLMs, each question reformulated 10 times in both Spanish and English, resulting in 2400 responses. Two anaesthesiologists were involved in the assessment process. One expert performed the initial ratings, while the second independently verified the evaluations assessing the outputs using an extended SERVQUAL framework. RESULTS: ChatGPT performed best, followed by Gemini 2. Medium-sized models, such as Phi-3 and OpenChat, outperformed several larger models like Llama-2 or Llama-3, challenging the notion that "bigger is better" and offering potential advantages in low-resource settings (e.g., Phi-3 outperformed Llama-2 with an average increase of 81% across all metrics). Specialized models did not show superior performance. Except for ChatGPT, English responses were generally more reliable than Spanish, with some Spanish outputs incoherent. ChatGPT also exhibited the least variability between responses. CONCLUSIONS: Despite promising performance, LLMs display limitations in medical contexts. Collaboration between national and international medical societies is crucial to develop evidence-based resources to guide LLM training and improve information trustworthiness.
10 May 2026
Read appraisal →Medical image analysis
Transfer learning from 2D natural images to 4D fMRI brain images via geometric mapping
Functional magnetic resonance imaging (fMRI) allows real-time observation of brain activity through blood oxygen level-dependent (BOLD) signals and is extensively used in studies related to sex classification, age estimation, behavioral measurements prediction, and mental disorder diagnosis. However, the application of deep learning techniques to brain fMRI analysis is hindered by the small sample size of fMRI datasets. Transfer learning offers a solution to this problem, but most existing approaches are designed for large-scale 2D natural images. The heterogeneity between 4D fMRI data and 2D natural images makes direct model transfer infeasible. This study proposes a novel geometric mapping-based fMRI transfer learning method that enables transfer learning from 2D natural images to 4D fMRI brain images, bridging the transfer learning gap between fMRI data and natural images. The proposed Multi-scale Multi-domain Feature Aggregation (MMFA) module extracts effective aggregated features and reduces the dimensionality of fMRI data to 3D space. By treating the cerebral cortex as a folded Riemannian manifold in 3D space and mapping it into 2D space using surface geometric mapping, we make the transfer learning from 2D natural images to 4D brain images possible. Moreover, the topological relationships of the cerebral cortex are maintained with our method, and calculations are performed along the Riemannian manifold of the brain, effectively addressing signal interference problems. The experimental results based on the Human Connectome Project (HCP) dataset demonstrate the effectiveness of the proposed method. Our method achieved state-of-the-art performance in sex classification, age estimation, and behavioral measurement prediction tasks. Moreover, we propose a cascaded transfer learning approach for depression diagnosis, and proved its effectiveness on 23 depression datasets. In summary, the proposed fMRI transfer learning method, which accounts for the structural characteristics of the brain, is promising for applying transfer learning from natural images to brain fMRI images, significantly enhancing the performance in various fMRI analysis tasks.
10 May 2026
Read appraisal →Neural networks : the official journal of the International Neural Network Society
Object-centric proto-symbolic behavioural reasoning from pixels
Autonomous intelligent agents must bridge computational challenges at disparate levels of abstraction, from the low-level spaces of sensory input and motor commands to the high-level domain of abstract reasoning and planning. A key question in designing such agents is how best to instantiate the representational space that will interface between these two levels-ideally without requiring supervision in the form of expensive data annotations. These objectives can be efficiently achieved by representing the world in terms of objects (grounded in perception and action). In this work, we present a novel, brain-inspired, deep-learning architecture that learns from pixels to interpret, control, and reason about its environment, using object-centric representations. We show the utility of our approach through tasks in synthetic environments that require a combination of (high-level) logical reasoning and (low-level) continuous control. Results show that the agent can learn emergent conditional behavioural reasoning, such as (A → B)∧(¬A → C), as well as logical composition (A → B)∧(A → C)⊢A → (B∧C) and XOR operations, and successfully controls its environment to satisfy objectives deduced from these logical rules. The agent can adapt online to unexpected changes in its environment and is robust to mild violations of its world model, thanks to dynamic internal desired goal generation. While the present results are limited to synthetic settings (2D and 3D activated versions of dSprites), which fall short of real-world levels of complexity, the proposed architecture shows how to manipulate grounded object representations, as a key inductive bias for unsupervised learning, to enable behavioral reasoning.
9 May 2026
Read appraisal →Artificial intelligence in medicine
Learning with less: A survey of deep learning in medical imaging under varying supervision levels
The rapid evolution of deep learning has significantly advanced the field of medical image analysis. However, despite these achievements, further progress is hindered by the scarcity of large, well-annotated datasets. To overcome this limitation, recent years have seen a growing focus on developing methods that leverage varying levels of supervision as alternatives to fully supervised learning. Instead of relying solely on complete annotations, these approaches employ partial annotations or other supervision strategies to train deep learning models. This paper presents a systematic review of deep learning methods for medical image analysis under different levels of supervision. To this end, we categorize these methods based on the level of supervision they rely on, encompassing categories such as no supervision, inexact supervision, incomplete supervision, inaccurate supervision, and only limited supervision. We further divide these categories into finer subcategories. For example, we categorize inexact supervision into multiple instance learning and learning with weak annotations. Similarly, we categorize incomplete supervision into semi-supervised learning, active learning, and domain-adaptive learning and so on. Furthermore, we systematically summarize commonly used datasets for different types of supervision in deep learning for medical image analysis and explore future research directions to conclude this survey.
9 May 2026
Read appraisal →Therapeutic innovation & regulatory science
Applications and Challenges of Artificial Intelligence and Big Data in Drug Regulation
Artificial intelligence (AI) and big data are increasingly applied in drug regulation and have demonstrated significant potential worldwide. The U.S. Food and Drug Administration (FDA) has developed a relatively comprehensive approach through strategic frameworks, regulatory guidelines, and pilot programs. The European Medicines Agency (EMA) has promoted AI adoption via the Big Data Task Force, DARWIN EU®, and a multi-annual work plan, while Japan, Canada, and other countries have also advanced relevant initiatives and strengthened international cooperation. In China, smart regulation has been incorporated into the "14th Five-Year Plan" and subsequent strategies, with progress in establishing national regulatory data platforms, pharmaceutical traceability systems, and pilot AI applications. Nevertheless, AI in drug regulation remains at an exploratory stage, facing challenges such as limited model reliability and interpretability, insufficient data standards and interoperability, regulatory gaps, and ethical as well as public trust concerns. Future progress will depend on strengthening regulatory standards, enhancing data governance, improving regulatory capacity, and deepening international collaboration to achieve more scientific, intelligent, and efficient drug regulation.
9 May 2026
Read appraisal →Neural networks : the official journal of the International Neural Network Society
Semi-supervised classification and projection with adaptive flexible structure optimal graph
Graph-based semi-supervised learning (GSSL) has received more attention in recent years. Many existing methods obtain a fixed similarity graph based on the original data. Affected by the redundant information and noise of the original data, the obtained similarity graph is not optimal, which greatly affects the subsequent work. And the similarity graphs obtained by these methods are fully connected, so the local structure information of the data cannot be preserved. Moreover, many GSSL methods project the original data into a low-dimensional space through a linear mapping function. However, linear functions may not be appropriate for data embedded in a nonlinear manifold. In order to overcome these shortcomings mentioned above, we relax the linear mapping function and impose a ℓ0-norm constraint on the similarity graph to gain an adaptive flexible structure optimal graph. Furthermore, incorporating the principle of maximum separability, an efficient GSSL method is proposed, named semi-supervised classification and projection with adaptive flexible structure optimal graph (SAFSG). Combining the construction of similarity graphs and label propagation, SAFSG can simultaneously get an adaptive flexible structure optimal graph, a label prediction matrix and a projection matrix. In addition, an efficient iterative algorithm for optimizing SAFSG is proposed. Finally, we conduct experiments on more than ten benchmark datasets, and the experimental results show that SAFSG performs satisfactorily in both classification and projection.
9 May 2026
Read appraisal →Journal of biomedical informatics
Model utility and explainability in federated learning - A case study in healthcare using fundus oculi datasets
OBJECTIVE: Introduce a case study for Federated Learning (FL) in healthcare, addressing challenges posed by patient privacy and limited large-scale datasets. Our goal is to assess the features learned by FL methods in a simulated, diverse setting that emphasizes realistic data heterogeneity, and to analyze the learned representations for their medical relevance using both local and global explainability techniques. METHODS: Six fundus oculi datasets were combined to simulate a diverse federated learning environment, representing heterogeneous data conditions. We evaluated three established FL methods against centrally trained models, assessing both predictive performance and the learned representations. Specifically, explainability techniques were employed to examine the features learned by the models, and local explanations were evaluated against attention maps annotated by ophthalmologists. Robustness against common biases in fundus datasets was also assessed. RESULTS: Our study found improvements in model utility (up to 9.97%) with FL methods compared to isolated training. Analysis of learned representations revealed that federated models predominantly learn the vertical cup-to-disc ratio, a crucial feature for glaucoma diagnosis, and demonstrated robustness against common biases. High agreement was observed between local explanations and ophthalmologist-annotated attention maps. CONCLUSION: This study demonstrates the benefits of FL systems in a healthcare scenario, providing a case study for evaluating federated systems beyond idealized benchmarks. Our findings highlight the potential of FL to not only improve model utility in privacy-sensitive medical domains but also to learn medically relevant features instead of spurious correlations.
9 May 2026
Read appraisal →Graefe's archive for clinical and experimental ophthalmology = Albrecht von Graefes Archiv fur klinische und experimentelle Ophthalmologie
Comparing the performance of four mainstream large language models on medical literature review generation: a human expert evaluation in SMILE surgery
PURPOSE: To systematically evaluate and compare the performance of four leading large language models (LLMs) in generating medical literature reviews across topics of varying research maturity, thereby providing insights for their effective and responsible application in academic writing. METHODS: In this comparative study, using standardized prompts, we instructed four leading LLMs (GPT-4, Gemini 2.5 Pro, Grok-3, and DeepSeek R1) to generate literature reviews on nine topics related to small incision lenticule extraction (SMILE) surgery. These topics were categorized into three groups by research maturity: well-researched, controversial, and open. Seven ophthalmology experts evaluated the generated content across four dimensions: quality, accuracy, bias, and relevance, while all references were verified for authenticity. Performance differences among models were evaluated using group comparison tests followed by post-hoc analysis. RESULTS: Significant performance variations were identified across all four models and dimensions (p < 0.001). Specifically, Gemini ranked highest in content quality, accuracy, and bias control. In contrast, DeepSeek, despite its high-quality score, received the lowest relevance score. Grok-3 demonstrated the highest reference authenticity (p < 0.001), whereas GPT-4's was the lowest (p < 0.001). All models showed diminished performance on open topics and exhibited severe reference fabrication ("hallucinations"). CONCLUSION: Rather than excelling universally, LLMs exhibit distinct and task-specific strengths that mandate a task-driven, hybrid strategy in tool selection. Reference fabrication was found to be a pervasive issue across all models, regardless of the task topic, elevating human verification from a best practice to an essential safeguard for academic integrity.
9 May 2026
Read appraisal →Radiology. Imaging cancer
ONCO-RADS-guided Large Language Models for Extraction and Classification of Incidental Findings on Whole-Body Imaging Reports
Purpose To evaluate large language model (LLM)-based strategy performance for extraction and classification of incidental findings from whole-body (WB) imaging reports, particularly strategies incorporating Oncologically Relevant Findings Reporting and Data System (ONCO-RADS). Materials and Methods In this retrospective bicenter study, authors included all WB MRI reports from January 2016 to December 2023 at a referral center (internal dataset). Two observers extracted all incidental findings, and patient records were used to confirm final diagnoses. First, authors evaluated ONCO-RADS performance and the reproducibility of its incidental finding classifications by six radiologists. Then, authors evaluated the accuracy of three LLM-based strategies: (a) a fine-tuned DeBERTa/medical named entity recognition (NER) model; (b) zero-shot LLMs (ChatGPT-o1 [OpenAI], Gemini-2.5-Pro [Google]); and (c) reference-guided prompting of these LLMs using ONCO-RADS. Authors then expanded these strategies to an external dataset of 605 reports with multiple imaging techniques (405 WB MRI; 100 fluorodeoxyglucose PET/CT; and 100 chest-abdomen-pelvis CT acquisitions) from January 2022 to January 2025. Results The internal dataset included 823 patients (mean age, 63.7 years ± 11.7 [SD]; 457 male patients) with 1488 WB MRI reports. The average interobserver reproducibility of ONCO-RADS incidental finding classifications was excellent (Cohen κ, 0.87). The per-report accuracies of ONCO-RADS-guided LLMs (95.6% [151 of 158] and 86.7% [137 of 158] for ChatGPT-o1 and Gemini-2.5-Pro, respectively) were higher than those of the medical NER (69.0% [109 of 158]) and zero-shot LLMs (57.0% [90 of 158] and 70.9% [112 of 158] for ChatGPT-o1 and Gemini-2.5-Pro, respectively) (P < .001). In the external test set (mean age, 60.6 years ± 12.9; 330 male patients), the per-report accuracies of ONCO-RADS-guided ChatGPT-o1 (83.5% [505 of 605]) and Gemini-2.5-Pro (82.0% [496 of 605]) were higher than those of the models without ONCO-RADS prompting (63.1% [382 of 605] and 61.2% [370 of 605], respectively) and the medical NER (55.7% [337 of 605]) (P < .001). Conclusion Reference-guided prompting of the LLMs ChatGPT-o1 and Gemini-2.5-Pro with ONCO-RADS improved their performance in extracting and classifying incidental findings on WB imaging reports compared with zero-shot prompting and medical NER. Keywords: Large Language Models, Incidental Findings, Whole-Body MRI Supplemental material is available for this article. © RSNA, 2026.
6 May 2026
Read appraisal →Artificial intelligence in medicine
TADynFed: Dynamic modality-adaptive federated learning with tissue-aware disentanglement for cross-disease analysis
Federated learning (FL) enables collaborative medical image analysis across decentralized institutions while preserving data privacy. However, real-world deployment faces critical challenges: modality heterogeneity, where clients possess incomplete or varying medical image modalities, and cross-disease generalization, requiring models to adapt to unseen pathologies with anatomical consistency. Existing methods like FedAvg, FedProx, IOP-FL, and PntTranForFL assume uniform modality availability and static client participation, leading to poor performance under realistic clinical constraints. We propose TADynFed, a novel framework for the Heterogeneous Federated Learning (HFL) paradigm, which addresses both data and modality heterogeneity through, a tissue-aware disentanglement strategy that decouples modality-tailored and modality-shared features, a dynamic prototype memory bank for missing modality compensation, an adaptive aggregation mechanism that accounts for client reliability and tenure. We evaluate TADynFed using multi-disease MRI datasets including BraTS21 along with cross-domain imaging data from CheXpert (chest X-ray) and Hep-2 (microscopy), simulating a 13-client FL environment. The proposed framework achieves an average mDice score of 66.03%, significantly outperforming baseline methods such as PointTransformerFL at 58.60% and FedAvg at 53.06%. It also attains the lowest boundary alignment error with an ASD of 1.85 mm and HD95 of 8.70 mm, surpassing existing approaches by notable margins. In terms of calibration stability, TADynFed records an ECE of 0.09, indicating superior confidence reliability. Furthermore, it demonstrates high communication efficiency with only 76 MB of data exchanged per round, compared to 95-105 MB in other frameworks. These results validate TADynFed's ability to maintain high segmentation accuracy, boundary precision, and calibration stability while minimizing bandwidth usage. Outperforming existing frameworks by significant margins, TADynFed demonstrates robust boundary alignment, superior calibration, and efficient communication. It also exhibits strong cross-disease transferability without retraining. By integrating structured representation decomposition, prototype-guided fusion, and client-adaptive learning, TADynFed establishes a new benchmark in realistic, heterogeneous, and mix-modal federated medical imaging systems. Code supporting this study TADynFed.
5 May 2026
Read appraisal →Annual review of biomedical engineering
Digital Twins for Biofluids
Digital twins-virtual representations dynamically linked to physical systems-have the potential to transform biomedical engineering by enabling real-time prediction, optimization, and personalization in health and disease. In biofluids, digital twins offer a framework for integrating physics-based models with data from clinical imaging, sensors, and physiological measurements to support diagnostics, therapeutic planning, and device design. This article reviews modeling approaches used in the construction of digital twins for biofluid applications. We survey high-fidelity numerical methods alongside emerging machine learning techniques, highlighting their respective strengths and limitations. Key requirements for digital twins are discussed, emphasizing the bidirectional interaction between physical and virtual assets and the importance of selecting modeling strategies tailored to specific biomedical contexts. While notable progress has been made over the past decade, significant challenges remain, particularly in integrating multiphysics models with data-driven methods and in establishing standardized protocols for data acquisition, interoperability, and sharing.
5 May 2026
Read appraisal →Science (New York, N.Y.)
Performance of a large language model on the reasoning tasks of a physician.
More than 65 years ago, complex clinical diagnostic reasoning cases were introduced as the gold standard for the evaluation of expert medical computing systems, a standard that has held ever since. In this study, we report the results of a physician evaluation of a large language model (LLM) on challenging clinical cases across five experiments with a baseline of hundreds of physicians. We then report a real-world study comparing human expert and artificial intelligence (AI) second opinions in randomly selected patients in the emergency room of a major tertiary academic medical center. In all experiments, the LLM outperformed physician baselines and displayed continued improvement from prior generations of AI clinical decision support. Our study suggests that LLMs have eclipsed most benchmarks of clinical reasoning, motivating the urgent need for prospective trials.
5 May 2026
Read appraisal →Microvascular research
OPU-NET-DADENA: Optimized deep learning ensemble with u-net segmentation for early detection of diabetic retinopathy
One of the most significant microvascular complications of diabetes mellitus (DM) is diabetic retinopathy (DR). In the early stages, patients with DR may not exhibit any noticeable symptoms. It is a diabetic disorder that damages retinal blood vessels in the eyes. At first, there are no symptoms or sporadic visual issues. When it worsens, it affects both eyes and can lead to partial or total blindness. A person who already has DM is therefore constantly at a higher risk of developing the condition. Early identification can prevent the possibility of total and irreversible blindness. Therefore, needs an efficient and early diagnosis system. So, this paper proposes a new deep-learning methodology in a specific Deep Siamese DenseNet for early detection of DR. The proposed method is accomplished through various steps, such as Data Collection, Preprocessing, Augmentation, Segmentation, Feature Extraction, Feature Selection, and Detection. An adaptive histogram equalization approach named Contrast Limited Adaptive Histogram Equalization (CLAHE) is used to preprocess input images, which reduces amplification of noise. Then, the segmentation is done by the Optimized U-NETs. Next, features of the contrasted retinal images are extracted by residual attention EfficientNet (RA-EfficientNet). Then, the optimal features are selected by a hybrid Reptile Search Algorithm (RORS). Finally, the deep learning methodology includes DarkNet, DenseNet 201, and NasNetMobile used to detect diabetic retinopathy at an early stage. The model is implemented in MATLAB and evaluated using accuracy, precision, F-score, specificity, sensitivity, MCC, NPV, FPR, and FNR. The proposed approach achieves 99.33% accuracy and 98.32% precision, outperforming previous methods that reported accuracies of 95-97% and precisions of 94-96%, demonstrating its effectiveness for reliable early detection of DR.
5 May 2026
Read appraisal →Journal of molecular graphics & modelling
DeepHybridCPI: A hybrid deep learning framework for compound-protein interaction prediction
In bioinformatics, deep learning-based methods for Compound-Protein Interaction (CPI) prediction play a vital role in virtual screening, drug discovery, and drug repositioning. Recent improvements in computational methods have shown great possibility to save costs of experiment and speed up target identification. Nevertheless, the current CPI forecasting methods remain severely limited. Many rely on shallow Graph Neural Networks (GNNs) that struggle to capture the global structural context of compounds, while conventional Convolutional Neural Networks (CNNs) focus primarily on local sequence motifs and fail to model long-range dependencies in proteins. Even though a number of recent architectures strive to solve these problems by adding complexity to models, or by adding complex modules, these additions often cause significant computational overhead. To overcome these challenges, we propose DeepHybridCPI, a hybrid deep learning framework designed for accurate and efficient CPI prediction. Our hybrid model integrates a multiscale, densely connected GNN to extract compound features capturing both local substructures and global molecular topology, and employs CNNs with Long Short-Term Memory (LSTM) networks to model both local motifs and extended dependencies in protein sequences. The learned compound and protein representations are fused into a unified latent space to enable effective interaction modeling. Experimental evaluations on benchmark Human and C. elegans datasets demonstrate that DeepHybridCPI consistently outperforms existing state-of-the-art baseline methods in terms of AUC, Precision, and Recall. These findings highlight the importance of combining multiscale compound representations with hybrid sequence encoders within a single unified framework, providing a promising avenue for accelerating computational drug discovery. We release our source code and dataset at: https://github.com/jamshaidwarraich/DeepHybridCPI.
4 May 2026
Read appraisal →International journal of neural systems
Data-related Ablation for Reinforcing Deep Learning in Explaining Complex Phenomena
Deep Learning (DL) models excel at automatically learning intricate patterns within complex data, but their black box nature undermines human trust. To address this, current validation strategies typically focus on the model itself, modifying its architecture to assess the role and importance of the components. However, this model-centric view overlooks the critical learning substrate, which is represented by the data, implicitly assuming that it accurately represents the target phenomenon. This implicit trust in data means that evaluation may fail to detect whether high performance stems from exploiting biases or data quirks rather than learning relevant patterns. We present a novel data-related ablation as a complement to the traditional architectural ablation. Using this framework for Electroencephalography (EEG) signals of Emotional Recognition (ER) and Motor Execution (ME) as a case study, we show that seemingly high-accuracy models often rely heavily on process-irrelevant features, maintaining performance even when key information is eliminated. This shows that a standard, data-independent evaluation can be misleading about whether a model truly captured the intended process; the proposed approach helps distinguish robust learning from leaning on incidental characteristics. Therefore, incorporating data-related ablation is essential for developing reliable and generalizable DL models in fields that rely on data derived from complex and often not completely known phenomena.
4 May 2026
Read appraisal →Scientific reports
Hybrid deep learning and YAMNet features for asthma diagnosis from respiratory sounds
A prompt and precise diagnosis is necessary to lessen the worldwide burden of asthma, a common chronic respiratory condition. Spirometry is one of the resource-intensive conventional methods that frequently misclassifies cases because of overlapping symptoms with other respiratory illnesses. The Asthma Detection Dataset v2, which is openly accessible and comprises actual audio recordings of respiratory noises, is used in this work to demonstrate a hybrid deep learning architecture for automated asthma detection. After applying thorough preprocessing, which includes denoising, segmentation, and augmentation, the suggested pipeline employs a dual-path feature extraction technique that combines pretrained YAMNet embeddings with handmade acoustic descriptors (MFCCs, chromagram, ZCR, and spectral features). Atrous Spatial Pyramid Pooling (ASPP) and Squeeze-and-Excitation (SE) modules are used to further improve these embeddings, and a Multi-Layer Perceptron is used for classification. The proposed model achieved high performance, with an accuracy of 98.6% ± 0.14, an F1-score of 97.5% ± 0.14, and a macro-AUC of 99.1% ± 0.14 using stratified 5-fold cross-validation. SHAP-based interpretability and visualization (waveforms, spectrograms, t-SNE) verified that the model records clinically significant auditory biomarkers such as crackles and wheezes. These results indicate the potential of the proposed model as an explainable and scalable computer-aided asthma diagnostic tool.
3 May 2026
Read appraisal →Turkish journal of ophthalmology
Large Language Models in Ophthalmology: A Bibliographic Analysis
This study evaluated the distribution of research on the use of large language models (LLMs) in ophthalmology through a bibliographic analysis of articles retrieved from PubMed through November 2024. Studies were categorized into four main areas of LLM application: clinical decision-making (further divided according to subspecialties), education, patient interactions, and miscellaneous applications. Descriptive statistics were used to analyze the distribution of studies by ophthalmic subspecialty, geographical region, journal quality, and author characteristics, including gender and scholarly impact (h-index and i10-index). The findings revealed that clinical decision-making was the most common application (43.7%), with the majority of studies in this subgroup focusing on the retina (39.5%). Geographically, most of the research originated from North America (48.3%), followed by Asia (29.9%) and Europe (20.7%). Most studies were published in high-impact journals (Q1 journals: 74.7%), particularly for those related to clinical decision-making in retina (80.0%), glaucoma (100%), and multiple subspecialties (87.5%). Gender disparities were evident across all author roles, with female authors accounting for only 29.9% of first authors, 25.3% of last authors, and 26.4% of corresponding authors. The results suggest a need for greater diversity in terms of gender and geographic representation in LLM research in ophthalmology to promote inclusive progress in the field. Bu çalışma, PubMed’de Kasım 2024’e kadar yayımlanmış göz hastalıkları alanında büyük dil modellerinin (BDM) kullanımı ile ilgili araştırmaların bibliyografik analizini amaçlamaktadır. Çalışmalar, klinik karar desteği (yandallara göre ayrılmış), eğitim, hasta etkileşimleri ve diğer uygulamalar olmak üzere BDM yöntemlerinin dört ana uygulama alanına göre kategorize edilmiştir. Tanımlayıcı istatistikler kullanılarak çalışmaların, oftalmoloji alt uzmanlıkları, coğrafi bölge, dergi kalitesi ve yazar özellikleri (cinsiyet ve bilimsel etki, h-endeksi ve i10-endeksi dahil) açısından dağılımı analiz edilmiştir. Bulgularımız, klinik karar desteğinin en yaygın uygulama olduğunu (%43,7) ve bu alt grup içerisinde çalışmaların çoğunun retina üzerine odaklandığını (%39,5) ortaya koymuştur. Coğrafi olarak, araştırmaların çoğu Kuzey Amerika’dan (%48,3) yayımlanmakta olup sırasıyla Asya (%29,9) ve Avrupa (%20,7) Kuzey Amerika’yı izlemektedir. Çoğu çalışma, özellikle de klinik karar desteği alt grubunda, retina (%80,0), glokom (%100) ve birden fazla alt uzmanlık dalını kapsayan çalışmalar (%87,5), yüksek etki faktörüne sahip dergilerde yayımlanmıştır (Q1 dergileri: %74,7). Cinsiyet dağılımındaki eşitsizlik, tüm yazar rollerinde belirgin olarak gözlemlenmiştir. Çalışmalarda kadın yazarlar, ilk yazarların yalnızca %29,9’unu, son yazarların %25,3’ünü ve sorumlu yazarların %26,4’ünü oluşturmuştur. Sonuçlar, göz hastalıkları alanındaki BDM araştırmalarında cinsiyet ve coğrafi temsilde daha fazla çeşitliliğe ihtiyaç duyulduğunu ve bu çeşitliliğin alandaki kapsayıcı ilerlemeyi teşvik edeceğini göstermektedir.
2 May 2026
Read appraisal →Journal of chemical information and modeling
TriPDCL: A Tri-Pathway Prototype-Driven Contrastive Learning Framework for Cross-Modality Single-Cell Integration
The rapid development of single-cell sequencing technologies has laid a solid foundation for multiomics data integration. A central objective in multiomics integration is to attenuate cross-omic discrepancies stemming from technical batch effects and inherent biological variability, thereby facilitating a unified latent representation that supports robust downstream analyses. However, the intrinsic high sparsity of single-cell data and pronounced intercellular heterogeneity pose greater challenges for joint analysis. We foreground cellular heterogeneity and present TriPDCL, a prototype-based contrastive learning framework, which not only achieves effective transfer of heterogeneity information through an iterative prototype-learning update mechanism but also, by leveraging learnable prototype centers, enables precise construction of reliable positive-negative sample pairs. It tackles two principal challenges, the alignment of cross-modal heterogeneity and the robust learning, in multiomics analysis. Subsequently, comparative assessments conducted on five data sets in contrast to seven representative methods illustrate the superiority of the proposed approach.
2 May 2026
Read appraisal →JMIR formative research
Retrieval-Augmented Generation Versus GPT-4o for Patient-Facing Gynecological Cancer Information: Quality Evaluation
Retrieval-augmented generation improved overall quality scores for patient-facing gynecological cancer information mainly through better source attribution.
28 Apr 2026
Read appraisal →BMJ health & care informatics
Omission and hallucination prevalence of clinical guidelines in diagnostic large language model outputs
OBJECTIVE: Meaningful assessments of how large language models (LLMs) incorporate clinical guidelines require large-scale testing over many queries. Here, we evaluate the prevalence of clinical guideline omissions and hallucinations in a large sample of diagnostic LLM outputs. METHODS: We used simulated case vignettes and zero-shot prompting to generate diagnostic outputs and rationales from GPT-4.1 and DeepSeek-V3. English case vignettes were created for hypercholesterolaemia and type-2 diabetes mellitus. Each vignette contained identical medical information, while sociodemographic characteristics varied in terms of sex, ethnicity and location. We calculated the prevalence of existing and hallucinated clinical guidelines in LLM outputs across disease, LLM and sociodemographic characteristics. RESULTS: We analysed a total of 12 197 LLM outputs, which quantifies three hazard areas: omissions (up to 97% for DeepSeek-V3 and 46% for GPT-4.1), hallucinations (up to 9%) and inconsistencies (guideline citation rate ranging from 0% to 78.39% across sociodemographic vignettes). Omission and hallucination rates were generally similar across vignettes with different sex or ethnicity data, yet were particularly sensitive to patient location. DISCUSSION: This study highlights significant variability in clinical guideline prediction across two different diseases, three different sociodemographic variables and two LLMs, even when the LLMs were instructed by identical prompts, establishing clinical guideline prediction in LLM outputs as a stochastic event. CONCLUSION: The stochastic nature of LLMs creates a unique challenge for evidence generation and clinical deployment. Being able to measure and capture this stochasticity within high-quality research designs will be a prerequisite to advancing the responsible deployment of LLMs in healthcare.
27 Apr 2026
Read appraisal →Biomedizinische Technik. Biomedical engineering
AI in healthcare: mapping Romania's transition from research to intellectual property & industry solutions
OBJECTIVES: Romania is aligning its healthcare AI ecosystem with the European Union's AI Act (2024/1689), through its own National AI Strategy. Despite a growing volume of research in medical AI, significant gaps remain in translating this output into patents and domestically developed high-risk medical devices. This study assesses Romania's research productivity, patenting activity, and commercial solutions in medical AI. METHODS: We conducted an analysis combining bibliometric data, patent information, and medical device databases, and analyzed 619 Romanian-authored articles on medical AI, 272 patent records, and identified domestic AI-powered software providers. RESULTS: Romania's publication output in medical AI has surged post-2018, primarily driven by academic institutions in Bucharest and Cluj. Patents are predominantly filed by multinationals, indicating a potential disconnect between research and industrial output. Six AI medical software solutions were identified, most under the low-risk classifications. CONCLUSIONS: While Romania is expanding its medical AI research, it faces barriers in converting academic output into innovation. The dominance of foreign corporations in patent filings, reliance on international funding for high-impact research, and the scarcity of domestically developed high-risk medical AI solutions highlight important gaps. Addressing these disparities is essential for national alignment in the AI medical innovation space.
26 Apr 2026
Read appraisal →PloS one
Evaluation of large language models in medical examinations: A scoping review protocol
INTRODUCTION: Standardized medical examinations, used to assess trainee clinical competencies, provide a rigorous means to verify LLM accuracy and reliability in medical contexts. Although current evaluations use these exams to test LLMs' clinical reasoning, significant performance variations occur across different clinical scenarios. Existing methods struggle to adapt to evolving research needs. This study synthesizes prior research on LLMs in medical exams, highlighting current limitations and proposing future research directions. METHODS AND ANALYSIS: The formulation of the protocol was guided by the standards set forth in the JBI Manual for Evidence Synthesis. Following the establishment of precise inclusion/exclusion criteria and search strategies, we will execute systematic searches in the PubMed and Web of Science Core Collection databases. The method encompasses literature review, data extraction, analytical frameworks, and process mapping. By employing this method, researchers maintain methodological rigor during the entire research process. ETHICS AND DISSEMINATION: This protocol describes a method for performing a scoping review. The investigation focuses on the organized synthesis and examination of previously published research. It does not include human/animal experimentation or sensitive data collection. Ethical approval is not required for this literature-based study.
24 Apr 2026
Read appraisal →PloS one
Artificial intelligence for monitoring hand hygiene compliance in healthcare settings: A scoping review
BACKGROUND: Hand hygiene is a fundamental measure for preventing healthcare-associated infections, yet traditional monitoring methods are significantly limited by the Hawthorne effect, high resource demands, and an inability to assess procedural quality. Artificial intelligence (AI) technology has emerged as a transformative, automated, and objective approach to address these long-standing challenges. OBJECTIVE: This scoping review sought to systematically map the existing evidence, technical pathways, performance metrics, and implementation challenges of AI for monitoring hand hygiene compliance in healthcare settings. METHODS: Following the Joanna Briggs Institute (JBI) methodological framework and PRISMA-ScR guidelines, we searched five major databases (PubMed, Scopus, Embase, Web of Science, and IEEE Xplore) for articles published between January 2000 and September 2025, supplemented by grey literature searching and backward citation tracking. Two reviewers independently screened records, assessed full-text reports for eligibility, and extracted data, which were synthesized using descriptive statistics and thematic analysis. RESULTS: Of 800 records identified through database and supplementary searches, 45 studies (2007-2025) were included. The primary technical pathways identified were computer vision (53.3%), wearable sensors (24.4%), Internet of Things-integrated systems (13.3%), and radar/radio frequency-based systems (8.9%). While computer vision achieved high accuracy (95%) in setting-specific ICU models, performance dropped to 56% in generalizable models. Wearable systems demonstrated portability but showed 5%-10% lower specificity than vision-based approaches. Most evidence is derived from small-scale technical validations, with a significant lack of formal fairness analysis and evaluation of clinical workflows or cost-effectiveness. CONCLUSION: AI-based hand hygiene monitoring shows promise for supporting more objective and scalable hand hygiene surveillance in healthcare settings. However, the field remains at a largely pre-translational stage. Future research should shift from technical feasibility toward implementation science, focusing on establishing standardized motion databases, evaluating ethical governance (e.g., privacy and automation bias), and conducting pragmatic trials to demonstrate sustained clinical benefit and organizational sustainability.
24 Apr 2026
Read appraisal →PloS one
DSCANet: Integrating dual encoder and spatial cross-attention for polyp segmentation
Colonoscopy is a crucial clinical procedure for detecting colorectal polyps, which are strongly associated with the development of colon cancer. This endoscopic technique plays a vital role in both cancer prevention and early diagnosis. Accurate and efficient polyp segmentation is critical for enhancing the diagnostic reliability and clinical utility of colonoscopy. However, achieving precise segmentation presents significant challenges, primarily due to the diversity of polyps in their size and shape, coupled with poorly defined boundaries between polyps and surrounding tissues. To address these challenges, we propose a novel segmentation network, named DSCANet, which is a dual-branch encoder-structured network designed to efficiently fuse body and edge features for high-precision medical image segmentation. DSCANet integrates four key modules: a dual-branch encoder, a spatial cross-attention (SCA) module, a bipolar fusion (BF) module, and a flexible axis-attention (FAA) module. The dual-branch encoder consists of separate body and edge encoders, which extract respective features independently. The SCA module bridges the semantic gap between the two encoders' features. The BF module fuses the shallowest and deepest features, while the FAA module assists the decoder in extracting semantic information from high-level features. DSCANet achieved superior performance on multiple colorectal polyp segmentation datasets. The code is available at https://github.com/Shantdst/DSCANet.
23 Apr 2026
Read appraisal →Scientific reports
End-to-end pipeline for automated heart failure diagnosis with clinical notes using SNOMED-CT
Diagnosis of heart failure is complex but crucial for patient outcomes and often hindered by the untapped potential of unstructured clinical notes. We introduce a novel end-to-end pipeline for heart failure diagnosis, leveraging electronic health records (EHR) and German clinical notes from 846 patients. Our pipeline synthesizes abbreviation disambiguation, translation of German clinical notes to English, medical entity linking to SNOMED-CT, and subsequent classification. The classification was performed using a Support Vector Machine (SVM) and compared against a fine-tuned medBERT.de neural baseline. We reduced the reliance on training data with zero-shot learning to address limitations with abbreviation disambiguation and entity linking approaches. Validation against benchmark datasets and cardiologists demonstrates high accuracy for real clinical use. Abbreviation disambiguation achieved an accuracy of up to 96.1%. Entity linking achieved competitive performance compared to state-of-the-art approaches on selected evaluation datasets. The SVM classification approach utilizing SNOMED-CT concepts and EHR data achieved an F1-score of 65.3%, on par with the medBERT.de neural baseline using clinical notes and EHR data. Despite challenges regarding limited language-specific resources and reference dataset availability for SNOMED-CT annotations in German, our pipeline demonstrates high potential for real-world clinical use and clinical decision support grounded in the standardized SNOMED-CT ontology.
21 Apr 2026
Read appraisal →PloS one
SCGAN-generated skin-tone-balanced extension of the HAM10000 dataset for fairer dermatological AI
This dataset presents a synthetic collection of skin lesion images generated using a tone-conditioned Generative Adversarial Network (GAN) to model skin tone variation in dermatological datasets, motivated by the limited representation of darker skin tones in commonly used benchmarks. The dataset includes 10,000 images, at a resolution of 128×128 pixels, equally divided between medium and dark skin tones, with lesion class labels assigned to follow the class distribution of the original HAM10000 dataset. To quantitatively characterize the generated data, we report Fréchet Inception Distance (FID) scores and a baseline classification experiment comparing models trained on HAM10000 alone versus HAM10000 augmented with the proposed synthetic images. This work provides a controlled synthetic extension intended to support fairness-aware experimentation, data augmentation, and downstream evaluation in dermatological machine learning.
20 Apr 2026
Read appraisal →Oncology reports
Incorporating artificial intelligence into morphological diagnosis of acute leukemias: Current landscape, challenges and prospects (Review)
Acute leukemias (ALs) are a diverse group of hematological malignancies characterized by the abnormal proliferation of immature cells. Microscopic observation of cell morphology based on the French‑American‑British classification remains a fundamental diagnostic method for ALs. However, manual screening from bone marrow smear images is often inefficient, laborious and prone to subjective bias, leading to potential misdiagnosis or missed diagnosis. Artificial intelligence (AI), particularly machine learning (ML), has expanded human capabilities in analyzing complex datasets, leading to breakthroughs in multiple fields, including medical research and clinical practice. Increasingly, ML applications are being developed to diagnose hematological diseases by extracting and aggregating morphological characteristics from peripheral blood and bone marrow smears. However, applying ML methods to recognize cell morphology in hematological diseases presents unique challenges compared with other pathology subspecialties. The present review provided an overview of AI and ML applications in ALs diagnosis, focusing on cell segmentation and data mining methods from microscopy images, and highlights their advantages over manual microscopy.
18 Apr 2026
Read appraisal →PloS one
Pretraining effective T5 generative models for clinical and biomedical applications
This paper presents a study of the impact of corpus selection and vocabulary design on the performance of T5-based language models in clinical and biomedical domains. We introduce five different T5-EHR models, each pretrained from scratch using different combinations of clinical and biomedical corpora alongside domain-specific vocabularies. We evaluated these models across a variety of clinical and biomedical tasks to quantify the impact of pretraining data and vocabulary tokenization choices on downstream performance. Our findings reveal the importance of aligning both pretraining corpus and vocabulary with the target domain. Models pretrained exclusively on clinical data achieve superior performance on clinical tasks, while adding biomedical data contributes only marginal gains in most cases, with a few exceptions. Similarly, the choice of vocabulary significantly influences model performance, with clinical-specific vocabularies outperforming general biomedical vocabularies in tasks requiring a deeper understanding of clinical language. Also, the T5 generative models perform competitively with state-of-the-art discriminative models on several biomedical benchmarks, demonstrating strong generalization to biomedical domain. Overall, these results emphasize that task-specific selection of corpus and vocabulary is essential for optimizing model performance in clinical and biomedical natural language processing (NLP).
18 Apr 2026
Read appraisal →Biomedical physics & engineering express
A multimodal attention fusion-based model for pathological voice detection
Pathological voice detection provides a non-invasive and cost-effective approach for the early screening of laryngeal disorders. However, most existing methods rely on single-modal acoustic features and mainly focus on binary classification, which limits their ability to achieve fine-grained discrimination among multiple pathological voice categories. To address these limitations, this paper proposes a Multimodal Fusion Network (MFNet) with attention-based fusion for pathological voice classification. The proposed framework jointly models time-domain and frequency-domain information within a unified architecture. Specifically, SincNet is employed to extract task-relevant narrowband acoustic features directly from raw speech waveforms, while the Tunable Q-factor Wavelet Transform (TQWT) and SKNet are combined to learn multi-scale spectral representations from Mel-spectrograms. In addition, an Attentional Feature Fusion (AFF) module is introduced to adaptively integrate features from the two branches, thereby enhancing the discriminative capability of pathological voice representations. Comparative experiments on three public pathological voice datasets show that the proposed method outperforms representative baseline models and demonstrates strong robustness and generalization across different datasets. Furthermore, ablation studies further validate the effectiveness of the proposed architectural design. These results indicate that MFNet provides an effective solution for pathological voice classification and has promising potential for computer-aided voice disorder screening.
18 Apr 2026
Read appraisal →International journal of medical informatics
Rule-augmented constraint learning for semantic error detection in MIMIC-III knowledge graph
High-quality, error-free data is essential for developing reliable data-driven models, particularly in clinical decision support systems where inaccurate predictions can have serious consequences. While KGs offer a structured and semantically rich representation for clinical data, ensuring their consistency and correctness remains a challenge. Existing rule mining techniques provide solutions for the automatic extraction of logical constraints from KGs, but they often produce redundant or clinically irrelevant rules, especially when dealing with numeric or categorical literals such as age or lab values. KG constraints-rules intended to capture implausible or conflicting facts in the KG-can be used to spot semantic errors: facts that might conform to the underlying schema but contradict domain knowledge. In this work, we propose a novel framework for constraint learning in clinical KGs that identifies and transforms high-confidence rules into clinically plausible constraints. We propose two approaches, based on class disjointness and literal clustering combined with rule mining. We validate the clinical relevance of these generated rules using expert-curated constraints and large language models (LLMs). The results on the MIMIC-III clinical dataset show that rule filtering based constraint learning effectively preserves clinically meaningful rules that align with established medical knowledge. For numeric data, we achieve reliable value groupings through our clustering-based method, and the rules derived from these groupings were validated by LLMs. Their outputs confirm the clinical relevance of a portion of those discovered rules. By providing interpretable and scalable solutions to semantic inconsistencies in KGs, this study contributes to increasing the KG trustworthiness and its clinical usability.
17 Apr 2026
Read appraisal →Computers in biology and medicine
Empirical evaluation of variability and multi-institutional generalizability of deep learning survival models: application to renal cancer CT scans
Deep learning survival (DLS) modeling has shown significant promise across multiple disease conditions for predicting patient outcomes via biomedical data, but remain relatively under-explored in conjunction with medical imaging. This poses the question of what methodological choices promote model generalization, especially when predicting continuous survival outcomes in external validation cohorts. In this work, we systematically examined the impact of data partitioning, data order, model initialization, and augmentation strategies on the generalization of CT-based DLS survival models in renal cancers. A multi-institutional CT cohort was assembled, leveraging public repositories such as TCGA-KIRC, TCGA-KICH, TCGA-KIRP, and KITS19; totaling 525 patients from across 9 institutions. A 3D ResNet-18 model was trained to predict overall patient survival with a Cox loss function and evaluated across three controlled experiments: (i) random versus covariate-balanced data partitioning, (ii) randomizing vs fixing data ordering and initial model weights, and (iii) varying intensity, spatial, and noise based augmentations; evaluated in terms of concordance index (c-index) and hazard ratio (HR). Intelligent covariate-balanced data partitioning demonstrated optimal performance on an external validation set (c-index 0.74, HR 5.08) compared to random partitioning (c-index 0.56, HR 0.29). Different model initializations significantly impacted the average performance and variance across replicate trainings, while data ordering had little to no effect. Combining multiple augmentations produced the best performance on external validation (c-index +4.76%, HR +44.39%) compared to no augmentations, while most impactful individual augmentation type was additive gaussian noise (c-index +3.17%, HR +23.47%). Our results demonstrate that careful selection of strategies for data partitioning and ordering, weight initialization, and image augmentation are critical for developing robust and generalizable DL models for continuous survival prediction using medical imaging data.
17 Apr 2026
Read appraisal →JMIR medical informatics
Classification of Cochrane Plain Language Summaries by Conclusiveness Using Transformer-Based Models and ChatGPT: Retrospective Observational Study
BACKGROUND: Cochrane plain language summaries (PLSs) aim to make systematic review findings more accessible to the general public. However, inconsistencies in how conclusions are presented may impact comprehension and decision-making. Classifying PLSs based on conclusiveness can improve clarity and facilitate informed health decisions. OBJECTIVE: This study aimed to develop and evaluate deep learning language models for the classification of PLSs according to 3 levels of conclusiveness (conclusive, inconclusive, and unclear) and to compare their performance with a general-purpose large language model (GPT-4o). METHODS: We used a publicly available dataset containing 4405 Cochrane PLSs of systematic reviews published until 2019, already classified by humans according to 9 categories of conclusiveness regarding the intervention's effectiveness or safety. We merged these categories into 3 classes based on the strength of conclusiveness: conclusive, inconclusive, and unclear. For the fine-tuning, we used Scientific Bidirectional Encoder Representations from Transformers (SciBERT), a pretrained language model trained on 1.14 million papers primarily from the health sciences, and Longformer, a transformer model designed specifically to process long documents. The script was developed using the Python programming language and the PyTorch framework. We computed evaluation metrics using the scikit-learn machine learning library and determined the area under the curve of the receiver operating characteristic (AUCROC) to measure the model performance in balancing sensitivity and specificity. We also analyzed a separate set of 213 PLSs and compared the predictions of our pretrained models with both manual verification and outputs generated by ChatGPT. RESULTS: The model based on SciBERT achieved a balanced accuracy of 56.6%. The AUCROC was 0.91 for "conclusive," 0.67 for "inconclusive," and 0.75 for "unclear" conclusiveness classes. The Longformer-based model had a balanced accuracy of 60.9%, with AUCROCs of 0.86 for "conclusive," 0.67 for "inconclusive," and 0.72 for "unclear" conclusiveness classes. Both models underperformed compared with ChatGPT, which demonstrated higher accuracy (74.2%), better precision and recall, and a higher Cohen κ (0.57). CONCLUSIONS: Fine-tuning 2 transformer-based language models showed mixed results in classifying Cochrane PLSs by conclusiveness, likely due to semantic overlap and subtle linguistic differences. Despite satisfactory internal test metrics, the fine-tuned models failed to generalize to newly published PLSs, where performance dropped to near-chance levels. These findings suggest that general-purpose large language models like GPT-4o may currently offer more reliable results for practical classification tasks in biomedical applications.
16 Apr 2026
Read appraisal →International journal of medical informatics
A structured decision-support framework for selecting imputation methods in clinical structured datasets: A secondary analysis
OBJECTIVE: Missing values are a common challenge in healthcare data analysis, and inadequate handling can introduce bias and undermine the validity of findings. Imputation methods offer a practical solution, but selecting an appropriate approach depends on multiple dataset-specific factors. This study proposes a structured decision-support framework that defines key prerequisites for choosing suitable imputation methods during the preprocessing of clinically structured datasets. METHODS: A secondary analysis of a previous systematic review was conducted, covering 69 studies to identify factors influencing imputation method selection. Domain experts evaluated assumptions regarding missing data characteristics and dataset structure, reaching consensus on the most relevant factors. These factors were synthesized into a structured framework designed to guide systematic and transparent imputation method selection in clinical data preprocessing workflows. RESULTS: Nine key factors were identified as essential for determining an appropriate imputation method. These include missing data characteristics, mechanism, pattern, and ratio and dataset attributes such as data type, variable role, distribution, and correlation. The ratio of missingness was the most influential factor, followed by variable role and missing value mechanism. Most studies emphasized the combined importance of both missing data properties and dataset features in imputation selection. CONCLUSIONS: Understanding the characteristics of missing values and dataset structure is crucial for selecting appropriate imputation methods. The proposed structured decision-support framework provides an evidence-based checklist to enhance transparency, reproducibility, and reliability in preprocessing clinical datasets within medical informatics workflows.
16 Apr 2026
Read appraisal →Computers in biology and medicine
MIPHEI-ViT: Multiplex immunofluorescence prediction from H&E images using ViT foundation models
Histopathological analysis is a cornerstone of cancer diagnosis, with Hematoxylin and Eosin (H&E) staining routinely acquired for every patient to visualize cell morphology and tissue architecture. On the other hand, multiplex immunofluorescence (mIF) enables more precise cell type identification via proteomic markers, but has yet to achieve widespread clinical adoption due to cost and logistical constraints. To bridge this gap, we introduce MIPHEI (Multiplex Immunofluorescence Prediction from H&E Images), a U-Net-inspired architecture that leverages a ViT pathology foundation model as an encoder to predict mIF signals from H&E images using rich pretrained representations. MIPHEI targets a comprehensive panel of markers spanning nuclear content, immune lineages (T cells, B cells, myeloid), epithelium, stroma, vasculature, and proliferation. We train our model using the publicly available OrionCRC dataset of restained H&E and mIF images from colorectal cancer tissue, and validate it on five independent datasets: HEMIT, PathoCell, IMMUcan, Lizard and PanNuke. On OrionCRC test set, MIPHEI achieves accurate cell-type classification from H&E alone, with F1 scores of 0.93 for Pan-CK, 0.83 for α-SMA, 0.68 for CD3e, 0.36 for CD20, and 0.28 for CD68, substantially outperforming both a state-of-the-art baseline and a random classifier for most markers. Our results indicate that, for some molecular markers, our model captures the complex relationships between nuclear morphologies in their tissue context, as visible in H&E images and molecular markers defining specific cell types. MIPHEI offers a promising step toward enabling cell-type-aware analysis of large-scale H&E datasets, in view of uncovering relationships between spatial cellular organization and patient outcomes.
15 Apr 2026
Read appraisal →Journal of medical Internet research
Error Detection in Emergency Radiology Reports Using a Large Language Model: Multistage Evaluation Study
BACKGROUND: Emergency radiology requires highly accurate reporting under time constraints; yet, increasing workloads raise the risk of errors. While large language models (LLMs) show potential for proofreading in general radiology, their performance in emergency settings and non-English contexts remains unclear. OBJECTIVE: We aim to evaluate the performance of a domain-optimized LLM, DeepSeek-R1, for identifying errors in Chinese emergency radiology reports, with comparison against assessments by board-certified radiologists. METHODS: We compiled 7435 emergency reports (dataset 1; radiography, computed tomography, and magnetic resonance imaging) collected from November 2024 to April 2025. In stage 1, a total of 5 LLMs were evaluated using 200 reports. The best model, DeepSeek-R1, proceeded to stages 2 and 3, where 0-shot and few-shot learning were tested on a separate set (n=100). Model performance was compared against 12 radiologists. Stage 4 validated real-world utility on 800 verified reports. RESULTS: In subdataset 1, under stress-testing conditions, DeepSeek-R1 achieved a higher error detection rate in the few-shot setting than in the 0-shot setting (84.4% vs 60.9%, P=.003). Its performance exceeded that of radiology residents (84.4% vs 51.6% and 53.1%, respectively; both P<.05) and showed no statistically significant difference compared with senior radiologists and attending radiologists (84.4% vs 68.8%-93.8%, P=.26 to ≥.99). Compared with residents, DeepSeek-R1 detected more critical omissions (100% vs 25% and 50%; both P<.05) and other errors (92% vs 33% and 33%; both P=.02). In dataset 2, collected from independent institutions, DeepSeek-R1 achieved a detection rate of 95% under the few-shot setting. Reading time was shorter than that of human readers (92 vs 109 s). In real-world validation, DeepSeek-R1 identified 117 true reporting errors, yielding a positive predictive value of 56.5%. CONCLUSIONS: DeepSeek-R1 holds promise for improving quality control in emergency radiology reports. Its performance and efficiency support its use as an assistive proofreading tool in real-world radiology workflows.
15 Apr 2026
Read appraisal →JMIR aging
Web-Based Personalized Machine Learning Recommendations to Enhance Shared Decision-Making in Prostate-Specific Antigen Screening: Randomized Controlled Trial
BACKGROUND: Prostate‑specific antigen (PSA) screening involves complex trade‑offs between early detection and the risks of overdiagnosis. For older adults (aged ≥50 years), shared decision‑making (SDM) is often hindered by limited health literacy, sensory or cognitive impairments, and multimorbidity, which complicate risk comprehension. Traditional decision aids provide foundational knowledge but are often nonpersonalized. Machine learning (ML) may offer individualized recommendations, yet the psychological and behavioral effects of ML‑assisted SDM in geriatric populations remain poorly characterized. OBJECTIVE: This study aimed to develop and evaluate a web‑based, ML‑driven decision aid integrated into an SDM workflow to provide personalized PSA screening recommendations and to assess its effects on decisional conflict (primary outcome), state anxiety, and decision satisfaction among middle‑aged and older men. METHODS: The study followed a 2‑stage design. First, a model establishment group (n=507) was used to train and evaluate 6 ML algorithms based on clinical and values‑clarification data. A random forest model was selected for its superior performance (mean area under the curve 0.933, SD 0.350; 95% CI 0.902-0.963). Second, a randomized controlled trial was conducted with 367 participants (mean age 64.34, SD 10.30 years) randomly assigned 1:1 to the ML suggestion group (MLSG; n=185) or the control group (CG; n=182). Both groups received video‑based education, counseling, and values clarification; only the MLSG received an ML‑generated "second opinion" recommendation. Primary and secondary outcomes were assessed using the Decisional Conflict Scale (DCS), Spielberger State‑Trait Anxiety Inventory (STAI), and Satisfaction with Decision scale. RESULTS: In the randomized controlled trial (n=367), the MLSG reported significantly lower decisional conflict than the CG (total DCS score: mean difference [MD] -3.77, 95% CI -5.55 to -1.99; Cohen d=-0.44; P<.001). The MLSG reported greater perceived support (DCS7: adjusted P=.03), more adequate advice (DCS9: adjusted P<.001), and higher decision confidence (DCS10: adjusted P=.03; DCS11: adjusted P<.001). Regarding psychological well‑being, although total anxiety scores did not differ, the MLSG reported reduced worry (STAI item 6: MD -0.98, 95% CI -1.20 to -0.76; d=-0.89; adjusted P<.001) and increased calmness (STAI item 1: MD 0.30, 95% CI 0.06-0.54; d=0.25; adjusted P=.01). Decision satisfaction was higher in the MLSG across all items (total Satisfaction with Decision score: MD -7.38, 95% CI -8.54 to -6.18; P<.001). Behavioral choices were strongly influenced by the ML recommendation: participants in the MLSG who received an "accept" recommendation were more likely to select "accept" (34/67, 50.7%) than those in the CG (44/182, 24.2%; P<.001). When the system suggested "not now," only 17.8% (21/118) chose "accept," which was lower than in the CG. CONCLUSIONS: Integrating personalized ML recommendations into SDM workflows provides emotional scaffolding for older men, reducing decisional distress and enhancing confidence without undermining autonomy. By addressing geriatric‑specific vulnerabilities through a facilitated digital interface, this ML‑driven approach complements traditional clinical consultations. These findings support the scalable integration of artificial intelligence-assisted decision support to foster patient‑centered care in aging populations.
15 Apr 2026
Read appraisal →Journal of medical Internet research
Context-Aware Sentence Classification of Radiology Reports Using Synthetic Data: Development and Validation Study
BACKGROUND: Automated structuring of radiology reports is essential for data utilization and the development of medical artificial intelligence models. However, manual annotation by experts is labor-intensive, and processing real clinical data through commercial large language models (LLMs) presents significant privacy risks. These challenges are particularly pronounced for non-English languages like Japanese, where specialized medical corpora are scarce. While synthetic data generation offers a potential privacy-preserving alternative, its effectiveness in capturing complex clinical nuances-such as negation and contextual dependencies-to train robust classification models without any real-world training data has not been fully established. OBJECTIVE: This study aimed to develop a context-aware sentence classification model for Japanese radiology reports using an entirely synthetic training pipeline, thereby eliminating reliance on real-world clinical data during the development phase. Furthermore, we sought to evaluate the generalizability of this approach by validating the model's performance on diverse, multi-institutional, real-world reports. METHODS: Japanese radiology reports (n=3104) were generated using GPT-4.1 and automatically annotated at the sentence level into 4 categories (background, positive finding, negative finding, and continuation) using GPT-4.1-mini. The synthetic data were partitioned into training (n=2670), validation (n=334), and test (n=100) sets. We fine-tuned several models, including lightweight local LLMs (Qwen3 and Llama 3.2 series) using low-rank adaptation and Japanese text classification models (Bidirectional Encoder Representations from Transformers [BERT]-base Japanese v3, Japanese Medical Robustly Optimized BERT Pretraining Approach [JMedRoBERTa]-base, and ModernBERT-Ja-130M). External validation was performed using 280 real-world reports (3477 sentences) from 7 institutions in the Japan Medical Image Database, with ground-truth labels established by board-certified radiologists. Evaluation metrics included accuracy, macro-averaged F1 (macro F1) score, and positive predictive value for positive findings (PPV_1). RESULTS: All models achieved high performance on the synthetic test set (accuracy: 0.938-0.951; macro F1-score: 0.924-0.940). Overall performance declined on the external validation dataset (accuracy: 0.783-0.813; macro F1-score: 0.761-0.790), reflecting distributional differences between synthetic and real-world reports; however, PPV_1 remained stable and high across datasets (eg, 0.957 on the synthetic test set vs 0.952 on the external validation dataset for Qwen3 [4B]). Parsing errors occurred in LLM-based approaches (19-260 sentences, 0.55%-7.48% in the external dataset). CONCLUSIONS: This study demonstrates the feasibility of developing context-aware sentence classification models for Japanese radiology reports using a training pipeline based entirely on synthetic data. The stability of PPV_1 indicates that the models successfully captured the essential clinical terminology and linguistic patterns required to identify positive findings in real-world reports, despite the observed performance degradation during external validation. This approach substantially reduces manual annotation requirements and privacy risks, providing a scalable foundation for constructing structured radiology datasets to support the development of clinically relevant medical artificial intelligence models.
11 Apr 2026
Read appraisal →Proceedings. Biological sciences
Statistical structure and the evolution of languages.
Human cultural development is marked by the emergence of new words and ideas, reflecting societal changes. But how does this evolution proceed? We use modern methods in natural language processing (namely, word embeddings) to measure statistical traces of cultural development, providing a testing ground to compare different models as to how this process works. We show that real embeddings of English and 21 other languages exhibit a series of previously unrecognized regularities. Specifically, these are: (i) frequency assortativity, where entities of high popularity cluster near other high-popularity entities; (ii) characteristic clustering velocity profiles due to aggregation into hierarchical structures; (iii) persistent temporal dynamics, where newly created entities appear disproportionately near other recent entries; and (iv) Taylor's law, implying that over time and across empirical semantic space the variance in new entity counts scales as a power of the mean, which helps systematize and quantify large historical fluctuations of neologisms. To explain these facts, we propose a class of generative models (specifically, directed preferential placement) that construct synthetic embeddings exhibiting similar regularities. We show that analogous regularities also occur in other datasets, suggesting that such generating models may shed light on new aspects of language and cultural evolution.
11 Apr 2026
Read appraisal →PloS one
MSCNet: Efficient and accurate semantic segmentation of LiDAR data using Multi-scale Convolution
In autonomous driving and intelligent robotics, the semantic information of LiDAR (Light Detection and Ranging) sensor data is crucial for understanding the surrounding environment. However, directly operating on point clouds is computationally expensive. To address this, some researchers have projected three-dimensional LiDAR data onto a two-dimensional spherical range view and used two-dimensional convolutional neural networks to segment the projected images. While the results are promising, many of these models are structurally complex, with high spatiotemporal complexity, which makes them unsuitable for real-time applications. To solve these issues, this paper proposes a multi-scale LiDAR data semantic segmentation method, MSCNet, with fewer parameters and higher segmentation accuracy. In the encoding phase, a single-channel multi-scale feature fusion block is introduced to alleviate the distribution differences between input channels. To obtain more stable local features, multi-scale dilated convolution residual blocks are designed to encode information from different receptive fields. To quickly capture global features, a pyramid pooling module is introduced. Experimental results on the SemanticKITTI, SemanticPOSS, and Pandaset datasets show that MSCNet achieves a good balance between parameter, accuracy, and running time. Particularly on the SemanticPOSS and Pandaset datasets, MSCNet achieves the best performance. Under the same parameter conditions, this method outperforms existing point cloud-based and projection-based methods.
10 Apr 2026
Read appraisal →Philosophical transactions. Series A, Mathematical, physical, and engineering sciences
The need for verification in artificial intelligence-driven scientific discovery
Artificial intelligence (AI) is transforming the practice of science. Machine learning (ML) and large language models (LLMs) can generate hypotheses at a scale and speed far exceeding traditional methods, offering the potential to accelerate discovery across diverse fields. However, the abundance of hypotheses introduces a critical challenge; without scalable and reliable mechanisms for verification, scientific progress risks being hindered rather than advanced. In this article, we trace the historical development of scientific discovery, examine how AI is reshaping established practices for discovery and review the principal approaches, ranging from data-driven methods and knowledge-aware neural architectures to symbolic reasoning frameworks and LLM agents. While these systems can uncover patterns and propose candidate laws, their scientific value ultimately depends on rigorous and transparent verification, which we argue must be the cornerstone of AI-assisted discovery. This article is part of the discussion meeting issue 'Symbolic regression in the physical sciences'.
10 Apr 2026
Read appraisal →Analytical chemistry
SIBioX: A Matrix Based Bioinformatics Analysis Tool Based on Swarm Intelligence Algorithm
Biological matrix data are essential for computational analysis, providing a structured framework to identify patterns and relationships in biological systems. Many other biological data types, including sequences, networks, and images, can be transformed into matrix representations through feature extraction and encoding. However, their high dimensionality complicates analysis, leading to increased computational complexity and the risk of overfitting, known as the curse of dimensionality. To address these challenges, we developed SIBioX, a matrix-based bioinformatics tool powered by swarm intelligence algorithms. It integrates 54 swarm intelligence methods, 5 conventional feature selection techniques, and 17 machine learning models, enabling comprehensive analysis of biological matrix data. With a user-friendly graphical interface, it supports operations such as feature normalization, selection, classification, clustering, statistical analysis, and data visualization. Additionally, it converts nonmatrix biological data, like gene and protein sequences, into matrix formats for further study. Experimental results demonstrate that SIBioX not only attains high accuracy in feature selection but also effectively reduces dimensionality, thereby streamlining bioinformatics workflows and promoting greater efficiency in biomedical research.
10 Apr 2026
Read appraisal →European thyroid journal
Promise and pitfalls of AI chatbots in complex decision-making for thyroid nodules and papillary thyroid cancer
INTRODUCTION: Artificial intelligence (AI) chatbots are increasingly used in medicine, but their reliability in scenarios with multiple management options is unclear. Indeterminate thyroid nodules and low- and low-to-intermediate-risk papillary thyroid carcinoma (PTC) represent such cases. METHODS: In a nationwide web-based survey, 201 members of the Hellenic Endocrine Society evaluated 12 clinical vignettes on indeterminate thyroid nodules and low- and low-to-intermediate-risk PTC. Their responses were compared with those generated by four conversational AI models (ChatGPT, Gemini, Copilot, and DeepSeek) at two time points, 11 months apart. DeepSeek was assessed only at the second time point. Chatbot outputs were assessed for agreement with endocrinologists' predominant answers, concordance with the most guideline-consistent options (American and European Thyroid Association recommendations), temporal stability, and inter-model agreement. RESULTS: Alignment between chatbots and endocrinologists' predominant responses was limited, reaching at most 25% across scenarios. In contrast, concordance with the most guideline-consistent options was higher, up to 83% (10/12 scenarios), depending on the model and time point. Across 12 scenarios, ChatGPT, Gemini, and Copilot changed their responses in 4, 7, and 5 scenarios, respectively, with some updates moving closer to, and others further from, guideline-based answers. Inter-model agreement ranged from 33 to 67%, indicating substantial variability among chatbots. CONCLUSION: AI chatbots show evolving but inconsistent performance in complex thyroid management scenarios. While guideline concordance can be relatively high, substantial variability across models, limited temporal reproducibility, and poor alignment with clinical practice highlight the need for ongoing longitudinal evaluation before safe integration into clinical decision-making.
9 Apr 2026
Read appraisal →Physics in medicine and biology
SSI-Net: a hybrid physics-constrained deep learning framework for quantitative ultrasound speed-of-sound reconstruction
Objective.Quantitative ultrasound tomography faces challenges in reconstructing speed-of-sound (SoS) distributions due to the ill-posed nature of the inverse problem and the computational complexity of full-waveform inversion. This study aims to develop an efficient and physically consistent deep learning framework for accurate SoS mapping from ultrasound signals.Approach. We propose speed of sound inversion network (SSI-Net), a dual data-physics-driven framework that integrates a bidirectional gated recurrent unit encoder for temporal feature extraction, a U-Net decoder for high-resolution spatial mapping, and a physics-constraint module based on an exact, differentiable finite-difference time-domain solver of the nonlinear Westervelt equation. The network model is trained through a joint optimization strategy that minimizes both data-fidelity loss and physics-informed residual loss, with a two-phase training schedule to ensure stable convergence.Main results.SSI-Net was validated across simulated, tissue-mimicking phantom, andin vivomouse datasets. It achieved superior reconstruction accuracy, with peak signal-to-noise ratio improvements of 7.4%-18.9% over state-of-the-art methods, while reducing the physics-based residual by 25.6%-57.5%. The framework demonstrated strong generalization to different transducer geometries (ring and linear arrays) and maintained stable performance across a ten-fold frequency range. Inference was completed within 109.4 ms per sample.Significance.By embedding an exact wave solver into a trainable architecture, SSI-Net combines the representational capacity of deep learning with strict physical consistency. It provides a robust, efficient, and clinically translatable solution for quantitative SoS imaging, offering a promising tool for tissue characterization in diagnostic ultrasound.
9 Apr 2026
Read appraisal →Journal of medical Internet research
Initial Insights Into an Institutional Secure Large Language Model for Magnetic Resonance Imaging Examination Requests: Retrospective Study
BACKGROUND: Incomplete clinical details on magnetic resonance imaging (MRI) examination requests (MERs) can lead to suboptimal protocol selection. An institutional secure large language model (sLLM) with access to manually retrieved salient data from the electronic medical record (EMR) may improve request completeness and protocol accuracy across multiple MRI subspecialties. OBJECTIVE: The objective of this study was to compare clinician MERs with sLLM-augmented MERs for information quality and to evaluate the protocoling accuracy of the sLLM versus board-certified radiologists across body, musculoskeletal, and neuroradiology MRI. METHODS: This retrospective study included 608 random outpatient MRI examinations performed between September 2023 and July 2024 (body 206, musculoskeletal 203, neuroradiology 199). The cohort comprised 528 patients (mean 51.2 years, SD 19.2; range 4-93; n=279, 52.8% women, n=249, 47.2% men). MERs without EMR access were excluded. A privately hosted Anthropic Claude 3.5 model (temperature 0) augmented each MER with manually retrieved salient EMR data and, via rule-based parsing, mapped the extracted elements onto predefined institutional criteria to recommend region or coverage and contrast use. Two experienced radiologists established a consensus reference standard. Two board-certified general radiologists (Rad 3 and Rad 4) and the sLLM were compared with this standard. Clinical information quality was graded using the Reason-for-Exam Imaging Reporting and Data System (RI-RADS). Interrater reliability was quantified with Gwet AC1. Paired accuracies were compared with the McNemar test to determine whether there was a statistically significant difference. RESULTS: Interreader agreement for RI-RADS was almost perfect for sLLM-augmented MERs (AC1 0.97, 95% CI 0.94-0.99) and moderate for clinician MERs (AC1 0.43, 95% CI 0.34-0.52). Limited or deficient clinical information (RI-RADS C/D) fell to 0% to 0.7% (0/608 to 4/608) with sLLM augmentation vs 4.1% to 20.4% (25/608 to 124/608) for clinician MERs. Overall protocol accuracy was 93.1% (566/608; 95% CI 89.6-96.6) for the sLLM, 91.4% (556/608; 95% CI 87.6-95.3) for Rad 3, and 92.1% (560/608; 95% CI 88.4-95.8) for Rad 4 (sLLM vs Rad 3 P=.23 vs Rad 4 P=.40). Region or coverage accuracy was similar (sLLM: 579/608, 95.2%; Rad 3: 585/608, 96.2%; Rad 4: 573/608, 94.2%; P=.46 and P=.36). Contrast decisions were more accurate using the sLLM at 94.4% (574/608; 95% CI 91.3-97.5) vs Rad 3 at 92.1% (560/608; 95% CI 88.4-95.8; P=.027) and were not significantly different to Rad 4 at 92.9% (565/608; 95% CI 89.4-96.4; P=.16). Subspecialty analyses showed similar patterns, with the sLLM outperforming Rad 4 for musculoskeletal MRI contrast decisions (96.6% vs 91.1%; P=.006) and matching readers elsewhere. Manual review indicated that sLLM improvements arose from EMR details not listed on the MER (infection/inflammation, tumor history, prior surgery). No clinically significant hallucinations were identified in a manual review of discordant cases. CONCLUSIONS: Across body, musculoskeletal, and neuroradiology MRI, sLLM-augmented examination requests improved clinical context and enhanced contrast selection while demonstrating accuracy comparable to general radiologists for region or coverage. Integrating sLLMs into routine vetting workflows may reduce manual workload in protocol selection for more efficient, standardized protocoling.
8 Apr 2026
Read appraisal →JMIR medical informatics
Deep Learning for Content-Based Medical Image Retrieval in Picture Archiving and Communication Systems for Brain Tumor Detection: Algorithm Development and Validation
BACKGROUND: Advances in medical imaging have led to massive archives, yet navigating these datasets remains challenging due to the limitations of traditional text-based search engines. While content-based medical image retrieval (CBMIR) offers a visual feature-based solution to enhance clinical workflows and research, its operational integration into picture archiving and communication systems (PACS) remains a significant bottleneck. Despite the progress in deep learning for feature extraction, CBMIR tools are rarely integrated and effectively implemented within existing radiology information systems due to complex protocol barriers. OBJECTIVE: To address these challenges, this study develops a CBMIR system meticulously designed to cater to 7 distinct types of brain tumors as seen in brain magnetic resonance images. Our system is tailored to assist radiologists and health care professionals in efficiently retrieving pertinent historical medical images, thereby providing quantitative decision support for radiologists and facilitating evidence-based case comparison, with the potential to improve retrieval efficiency and clinical workflow, rather than directly enhancing diagnostic accuracy. METHODS: The dataset used in this study was collected from a single medical center and is not publicly available. The core innovation is a state-of-the-art deep learning-based feature extraction algorithm specifically engineered for the CBMIR system. We use GoogLeNet as the primary architecture, incorporating generalized mean pooling to capture nuanced local features and an embedding layer for dimension reduction. Crucially, we address the integration gap by harmonizing 2 open-source projects to successfully embed the CBMIR system into a functional PACS environment via standard protocols. RESULTS: The image dataset contains 658 participants with 15,873 images collected from 2000 to 2017. The empirical findings of our research demonstrate the performance and robustness of the proposed CBMIR system. Our system achieves a remarkable mean average precision score of 89.16% and an equally impressive Precision@10 score of 94.08%. These metrics affirm the system's efficacy in retrieving relevant medical images. Furthermore, we successfully integrate the CBMIR system into a PACS by successfully harmonizing 2 open-source projects. CONCLUSIONS: This study presents the design and implementation of a PACS-integrated CBMIR system for brain magnetic resonance imaging, and the experimental results demonstrate that the system can achieve efficient and accurate image retrieval within a clinical workflow.
7 Apr 2026
Read appraisal →Genetics
Mondo: integrating disease terminology across communities
Precision medicine aims to enhance diagnosis, treatment, and prognosis by integrating multimodal data at the point of care. However, challenges arise due to the vast number of diseases, differing methods of classification, and conflicting terminological coding systems and practices used to represent molecular definitions of disease. This lack of interoperability artificially constrains the potential for diagnosis, clinical decision support, care outcome analysis, as well as data linkage across research domains to support the development or repurposing of therapeutics. There is a clear and pressing need for a unified system for managing disease entities-including identifiers, synonyms, and definitions. To address these issues, we created the Mondo disease ontology-a community-driven, open-source, unified disease classification system that harmonizes diverse terminologies into a consistent, computable framework. Mondo integrates key medical and biomedical terminologies, including Online Mendelian Inheritance in Man (OMIM), Orphanet, Medical Subject Headings (MeSH), National Cancer Institute Thesaurus (NCIt), and more, to provide a comprehensive and accurate representation of disease concepts with fully provenanced and attributed links back to the sources. Mondo can be used as the handle for curation of gene-disease associations utilized in diagnostic applications, research applications such as computational phenotyping, and in clinical coding systems in clinical decision support by pointing the clinician to the numerous knowledge resources linked to the Mondo identifier. Mondo's community-centric approach, stewarded by the Monarch Initiative's expertise in ontologies, ensures that the ontology remains adaptable to the evolving needs of biomedical research and clinical communities, as well as the knowledge providers.
5 Apr 2026
Read appraisal →European radiology
Deep learning-based segmentation of T1 and T2 cardiac MRI maps for automated disease detection.
OBJECTIVES: Parametric tissue mapping enables quantitative cardiac tissue characterization but is limited by inter-observer variability during manual delineation. Traditional approaches relying on average relaxation values and single cutoffs may oversimplify myocardial complexity. This study evaluates whether deep learning (DL) can achieve segmentation accuracy comparable to inter-observer agreement, explores the utility of statistical features beyond mean T1/T2 values, and assesses whether machine learning (ML) combining multiple intensity-based statistical features enhances disease detection. MATERIALS AND METHODS: T1 and T2 maps were manually segmented. The test subset was independently annotated by two observers, and inter-observer variability was assessed. A DL model was trained to segment the left-ventricular blood pool and myocardium. Average (A), lower quartile (LQ), median (M), and upper quartile (UQ) were computed for the myocardial pixels and employed in classification by applying cutoffs or in ML. Dice similarity coefficient (DICE) and mean absolute percentage error evaluated segmentation performance. Bland-Altman plots assessed inter-user and model-observer agreement. Receiver operating characteristic analysis determined optimal cutoffs. Pearson correlation compared features from model and manual segmentations. F1-score, precision, and recall evaluated classification performance. Wilcoxon test assessed differences between classification methods, with p < 0.05 considered statistically significant. RESULTS: 144 subjects (mean age 42.2 years ± 16.1, 76 men) were split into training (100), validation (15) and evaluation (29) subsets. Segmentation model achieved a DICE of 85.4%, surpassing inter-observer agreement. Random forest applied to all features increased F1-score (92.7%, p < 0.001). CONCLUSION: DL facilitates segmentation of T1/ T2 maps. Combining multiple features with ML improves disease detection. KEY POINTS: Question Manual segmentation of myocardial T1/T2 maps is time-consuming and affected by inter-observer variability; relying on single cutoff values for diagnosis may oversimplify myocardial complexity. Findings Deep learning achieves segmentation accuracy within inter-observer agreement, while machine learning improves disease detection compared to singular cutoffs. Clinical relevance Automated segmentation and feature extraction from T1/T2 maps can enhance workflow efficiency, reduce inter-observer variability, and improve diagnostic consistency. The high recall of the machine learning model minimizes missed diagnoses, ensuring more reliable disease detection.
4 Apr 2026
Read appraisal →Critical care clinics
Innovation, Technology, and Telemedicine in Critical Care
Critical care medicine is undergoing a major transformation driven by rapid technological innovation, digital integration, and telemedicine. This article examines how modern intensive care units (ICUs) are evolving into data-rich, precision environments equipped with advanced monitoring, organ support, and predictive analytics. Innovations such as artificial intelligence-enhanced decision support, portable ultrasound, and tele-ICU platforms are expanding diagnostic accuracy, remote access, and timely interventions. By tracing historical milestones to emerging technologies like smart ICUs and wearable biosensors, this article highlights the shift toward anticipatory, personalized care. It underscores the importance of evidence-based adoption to ensure these tools improve outcomes.
4 Apr 2026
Read appraisal →Pediatric radiology
Data mining in pediatric radiology in the era of artificial intelligence
Data mining is the systematic process of extracting useful knowledge from large multimodal datasets and is increasingly enabled by artificial intelligence (AI) methods. Pediatric radiology is a natural field for data mining because multimodal data sources, including images, reports, metadata, and electronic health records, together capture rich information on anatomy, disease, treatment, and outcomes. In the current era, the boundaries between data mining and AI are increasingly blurred. AI assists in key steps of the mining workflow through automated labeling, information extraction, and representation learning, while data mining provides the high-quality curated datasets that underpin model performance, generalizability, and safety. This review, therefore, examines both domains together, emphasizing their interdependence in the pediatric context. We describe core concepts and workflows of data mining in pediatric radiology, including data collection, linkage, annotation, analysis, validation, and governance, and outline how modern AI tools such as deep learning, large language models, multimodal fusion, and federated learning support advanced pattern discovery across limited and heterogeneous pediatric datasets. We summarize current and emerging clinical applications across diagnosis, prognosis, radiation dose monitoring, operational analytics, reporting safety nets, and continual learning. We then discuss current challenges related to data quality and standardization, ethics, regulation, workflow integration, resource disparities, sustainability, and explainability. Finally, we highlight future perspectives, including synthetic data generation, foundation models, structured reporting, and pediatric-focused ethical frameworks that aim to enable safe, transparent, and equitable integration of AI-driven data mining to improve outcomes in children.
4 Apr 2026
Read appraisal →Computer methods and programs in biomedicine
Application of artificial intelligence in colonoscopy imaging for polyp analysis-A systematic review
The colon is a major component of the digestive system, so early detection of colorectal polyps is essential in preventing colorectal cancer, which is a leading cause of cancer-related death worldwide. While colonoscopy remains the gold standard for polyp detection, its diagnostic accuracy is highly operator-dependent. Recent advances in Deep Learning (DL), a branch of Artificial Intelligence (AI), have shown substantial potential to improve colonoscopy image analysis by enhancing the accuracy, consistency, and objectivity of polyp detection, segmentation, and classification. Artificial intelligence-based systems have significantly reduced inter-observer variability and increased diagnostic efficiency, ultimately transforming the landscape of colorectal lesion assessment. This survey provides a comprehensive and critical analysis of the current status of deep learning applications in colorectal polyp analysis. We systematically review state-of-the-art methodologies across various DL architectures-including Convolutional Neural Networks (CNNs), transformer-based models, and hybrid approaches-and examine their performance on publicly available benchmark datasets. Additionally, we highlight the strengths and limitations of existing techniques, explore the clinical relevance of AI-assisted tools, and identify prevailing challenges such as data imbalance, real-time deployment, and generalizability across diverse populations and colonoscopy devices. By consolidating key advances and outlining future research directions, this review aims to serve as a valuable resource for researchers, clinicians, and developers seeking to leverage deep learning to enhance colorectal polyp detection, diagnosis, and clinical decision-making.
4 Apr 2026
Read appraisal →The Journal of pathology
Integrating artificial intelligence (AI) into colorectal cancer reporting
Artificial intelligence (AI) and deep learning (DL) are transforming cancer research and clinical care, with histopathology playing a central role in this transformation. In colorectal cancer (CRC), the second leading cause of cancer mortality world-wide, multimodal and vision-language models (VLMs) hold particular promise for enhancing the standardisation of histopathology reporting, the understanding of disease biology, and the discovery of novel prognostic indicators. Despite the availability of guidelines and reporting templates for essential prognostic indicators, variability remains in how key features such as TNM staging or tumour deposits are assessed and reported in routine clinical practice. AI-based tools have the potential to support refined extraction of established and extended features directly from whole-slide images. In parallel, recent studies have shown that DL models applied to pathology slides and associated AI-based biomarkers can outperform traditional histopathological prognostic indicators and uncover novel parameters, including tumour-adipocyte interactions, tumour-stroma ratio, and immune cell patterns at the invasive margin. Here, we review recent advances in both domains: AI-assisted standardisation of CRC pathology reporting and AI-driven identification of novel prognostic biomarkers. We highlight the need to refine and standardise CRC reporting practices and propose that a harmonised approach combining established pathology features with AI-derived prognostic indicators could refine risk assessment and improve outcomes for CRC patients. © 2026 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.
4 Apr 2026
Read appraisal →IEEE transactions on medical imaging
Self-Paced Learning for Images of Antinuclear Antibodies
Antinuclear antibody (ANA) testing is a critical method for diagnosing autoimmune disorders such as Lupus, Sjögren's syndrome, and scleroderma. Despite its importance, manual ANA detection is slow, labor-intensive, and demands years of training. ANA detection is complicated by over 100 coexisting antibody types, resulting in vast fluorescent pattern combinations. Although machine learning and deep learning have enabled automation, ANA detection in real-world clinical settings presents unique challenges as it involves multi-instance, multi-label (MIML) learning. In this paper, a novel framework for ANA detection is proposed that handles the complexities of MIML tasks using unaltered microscope images without manual preprocessing. Inspired by human labeling logic, it identifies consistent ANA sub-regions and assigns aggregated labels accordingly. These steps are implemented using three task-specific components: an instance sampler, a probabilistic pseudo-label dispatcher, and self-paced weight learning rate coefficients. The instance sampler suppresses low-confidence instances by modeling pattern confidence, while the dispatcher adaptively assigns labels based on instance distinguishability. Self-paced learning adjusts training according to empirical label observations. Our framework overcomes limitations of traditional MIML methods and supports end-to-end optimization. Extensive experiments on one ANA dataset and three public medical MIML benchmarks demonstrate the superiority of our framework. On the ANA dataset, our model achieves up to +7.0% F1-Macro and +12.6% mAP gains over the best prior method, setting new state-of-the-art results. It also ranks top-2 across all key metrics on public datasets, reducing Hamming loss and one-error by up to 18.2% and 26.9%, respectively. The source code can be accessed at https://github.com/fletcherjiang/ANA-SelfPacedLearning.
4 Apr 2026
Read appraisal →European radiology
Deep learning models for radiography body-part classification and chest radiograph projection/orientation classification: a multi-institutional study.
OBJECTIVES: Large-scale radiographic datasets often include errors in labels such as body parts or projection, which can undermine automated image analysis. Therefore, we aimed to develop and externally validate two deep-learning models-one for categorising radiographs by body part, and another for identifying projection and rotation of chest radiographs-using large, diverse datasets. MATERIALS AND METHODS: We retrospectively collected radiographs from multiple institutions and public repositories. For the first model (Xp-Bodypart-Checker), we included seven categories (Head, Neck, Chest, Incomplete Chest, Abdomen, Pelvis, Extremities). For the second model (CXp-Projection-Rotation-Checker), we classified chest radiographs by projection (anterior-posterior, posterior-anterior, lateral) and rotation (upright, inverted, left rotation, right rotation). Both models were trained, tuned, and internally tested on separate data, then externally tested on radiographs from different institutions. Model performance was assessed using overall accuracy (micro, macro, and weighted) as well as one-vs.-all area under the receiver operating characteristic curve (AUC). RESULTS: In the Xp-Bodypart-Checker development phase, we included 429,341 radiographs obtained from Institutions A, B, and MURA. In the CXp-Projection-Rotation-Checker development phase, we included 463,728 chest radiographs from CheXpert, PadChest, and Institution A. The Xp-Bodypart-Checker achieved AUC values of 1.00 (99% CI: 1.00-1.00) for all classes other than Incomplete Chest, which had an AUC value of 0.99 (99% CI: 0.98-1.00). The CXp-Projection-Rotation-Checker demonstrated AUC values of 1.00 (99% CI: 1.00-1.00) across all projection and rotation classifications. CONCLUSION: These models help automatically verify image labels in large radiographic databases, improving quality control across multiple institutions. KEY POINTS: Question This study examines how deep learning can accurately classify radiograph body parts and detect chest projection/orientation in large, multi-institutional datasets, enhancing metadata consistency for clinical and research workflows. Findings Xp-Bodypart-Checker classified radiographs into seven categories with AUC values of over 0.99 for all classes, while CXp-Projection-Rotation-Checker achieved AUC values of 1.00 across all projection and rotation classifications. Clinical relevance Trained on over 860,000 multi-institutional radiographs, our two deep-learning models classify radiograph body-part and chest radiograph projection/rotation, identifying mislabeled data and enhancing data integrity, thereby improving reliability for both clinical use and deep-learning research.
4 Apr 2026
Read appraisal →European radiology
Large language model-based uncertainty-adjusted label extraction for artificial intelligence model development in upper extremity radiography
OBJECTIVES: To evaluate GPT-4o's zero-shot ability to extract structured diagnostic labels (with uncertainty) from free-text radiology reports and to test how these labels affect multi-label image classification of musculoskeletal radiographs. MATERIALS AND METHODS: This retrospective study included radiography series of the clavicle (n = 1170), elbow (n = 3755), and thumb (n = 1978). After anonymization, GPT-4o filled out structured templates by indicating imaging findings as present ("true"), absent ("false"), or "uncertain." To assess the impact of label uncertainty, "uncertain" labels of the training and validation sets were automatically reassigned to "true" (inclusive) or "false" (exclusive). Label-image pairs were used for multi-label classification using the ResNet50 architecture. Label extraction accuracy was manually verified on internal (clavicle: n = 233, elbow: n = 745, thumb: n = 393) and external test sets (n = 300 for each). Performance was assessed using macro-averaged receiver operating characteristic (ROC) area under the curve (AUC), precision, recall curves, sensitivity, specificity, and accuracy. AUCs were compared with the DeLong test. RESULTS: Automatic extraction was correct in 98.6% (60,618 of 61,488) of labels in the test sets. Across anatomic regions, label-based model training yielded competitive performance measured by macro-averaged AUC values for inclusive (e.g., elbow: AUC = 0.80 (range, 0.62-0.87)) and exclusive models (elbow: AUC = 0.80 (range, 0.61-0.88)). Models generalized well on external datasets (elbow (inclusive): AUC = 0.79 (range, 0.61-0.87); elbow (exclusive): AUC = 0.79 (range, 0.63-0.89)). No significant differences were observed across labeling strategies or datasets (p ≥ 0.15). CONCLUSION: GPT-4o extracted labels from radiologic reports to train competitive multi-label classification models with high accuracy. Detected uncertainty in the radiologic reports did not influence the performance of these models. KEY POINTS: Question Can GPT-4o automatically extract high-accuracy, uncertainty-aware diagnostic labels from routine radiologic reports of the clavicle, elbow, and thumb for use in training multi-label image classifiers? Findings GPT-4o extracted labels with > 98% accuracy, and multi-label classifiers for clavicle, elbow, and thumb radiographs performed consistently regardless of how uncertainty was handled. Clinical relevance Automated GPT-4o-based labeling of routine clavicle, elbow, and thumb radiologic reports enables the rapid conversion of radiologic reports into structured multi-label training datasets, supporting scalable development of dedicated image classification models.
4 Apr 2026
Read appraisal →European journal of pediatric surgery : official journal of Austrian Association of Pediatric Surgery ... [et al] = Zeitschrift fur Kinderchirurgie
Latest Developments in Artificial Intelligence and Machine Learning Models in General Pediatric Surgery
Artificial intelligence (AI) and machine learning (ML) models rapidly transform health care with applications ranging from diagnostic image interpretation, predictive modeling, personalized treatment planning, real-time intraoperative guidance, and outcome prediction. However, their implementation in general pediatric surgery remains limited due to the rarity and complexity of pediatric surgical conditions, small and heterogeneous datasets, and a lack of formal AI training and competencies among pediatric surgeons.This narrative review explores the current landscape of AI and ML applications in general pediatric surgery, focusing on five key conditions: appendicitis, necrotizing enterocolitis, Hirschsprung's disease, congenital diaphragmatic hernia, and biliary atresia. For each, we summarize recent developments, including the use of AI in image analysis, diagnostic support, prediction of disease severity and outcome, postoperative monitoring, and histopathological evaluation. We also highlight novel tools such as explainable AI models, natural language processing, and wearable technologies.Recent findings demonstrate promising diagnostic and prognostic capabilities across multiple conditions. However, most AI/ML models still require external validation and standardization. The review underscores the importance of collaborative, multicenter research based on joint datasets as well as targeted AI education for pediatric surgeons to fully explore the benefits of these technologies in clinical practice.AI and ML offer significant potential to improve pediatric surgical care, but broader implementation will require multicenter collaboration, a robust dataset, and targeted AI education for pediatric surgeons.
4 Apr 2026
Read appraisal →Aesthetic Plast Surg
The Rise of Intelligent Plastic Surgery: A 10-Year Bibliometric Journey Through AI Applications, Challenges, and Transformative Potential
BACKGROUND: Driven by advancements in deep learning, surgical robots, and predictive modeling technologies, the integration of artificial intelligence (AI) and plastic surgery has expanded rapidly. Although AI shows the potential to enhance precision and efficiency, its clinical integration faces challenges, including ethical concerns and interdisciplinary complexity, which require a systematic analysis of research trends. METHODS: The CiteSpace and VOSviewer software were used to conduct a quantitative analysis of 235 documents in the core collection of Web of Science from 2016 to 2024. Co-citation networks, keyword co-occurrence, burst detection, and cluster analysis were employed to map the research trajectories. The inclusion criteria gave priority to studies that explicitly incorporated artificial intelligence into surgical designs or outcomes. The contributions of countries, institutions, and authors were evaluated through centrality indicators. RESULT: Publications related to artificial intelligence have grown exponentially, with the USA, Germany, and Canada leading research output. Harvard and Stanford Universities dominate in terms of institutional contributions, but cross-institutional collaboration remains limited. The keyword cluster highlights the innovations of artificial intelligence in breast reconstruction, facial analysis, and automated grading systems. Burst terms such as "deep learning," "risk assessment," and "attractiveness" underscore AI's role in optimizing surgical outcomes, but they also expose biases against Western-centric beauty standards. Ethical concerns, dataset diversity gaps, and overreliance on AI-driven decisions have become key obstacles. CONCLUSION: The integration of artificial intelligence in plastic surgery goes beyond the utility based on tools and into data-informed surgical engineering. The persistent gap in collaboration and dataset diversity highlights the need for global, interdisciplinary efforts to address technical and ethical challenges while advancing AI's clinical utility. Future research must prioritize transparency, inclusivity, and collaborative innovation to realize AI's transformative potential while mitigating risks. LEVEL OF EVIDENCE IV: This journal requires that authors assign a level of evidence to each article. For a full description of these Evidence-Based Medicine ratings, please refer to the Table of Contents or the online Instructions to Authors www.springer.com/00266 .
1 Mar 2026
Read appraisal →Int J Surg
Artificial Intelligence in melanoma research: a bibliometric analysis
BACKGROUND: As one of the most lethal skin cancers, melanoma has encountered many obstacles in diagnosis and therapy. Artificial intelligence (AI) can help improve early diagnosis, prognosis, and treatment of melanoma. However, there is a lack of detailed and accurate bibliometric analysis of the field. METHODS: All publications were extracted from Web of Science Core Collection based on AI and melanoma terms. Bibliometric analysis was conducted on 1476 articles/reviews by using VOSviewer, CiteSpace, and bibliometrix for co-authorship, citation, keyword, and journal analysis. RESULTS: There was a sudden increase in publications each year after 2017 and reached 260 in 2024. The United States (337 publications) and China (292 publications) ranked top in publication productivity. Germany was the top institution and author country. IEEE Access (57 publications) and Diagnostics (54 publications) were core journals. The two most prominent research hotspots were AI-assisted diagnosis and AI-integrated immunotherapy according to the keyword analysis. CONCLUSION: AI in melanoma research has exploded since 2017. It is recommended that future research focus on various datasets, explainable AI, and cross-disciplinary cooperation to promote the transformation of achievements into clinical practice.
1 Feb 2026
Read appraisal →Am J Perinatol
Application of Generative AI to Enhance Obstetrics and Gynecology Research
The rapid evolution of large-language models such as ChatGPT, Claude, and Gemini is reshaping the methodological landscape of obstetrics and gynecology (OBGYN) research. This narrative review provides a comprehensive account of generative AI capabilities, key use cases, and recommended safeguards for investigators. First, generative AI expedites hypothesis generation, enabling researchers to interrogate vast corpora and surface plausible, overlooked questions. Second, it streamlines systematic reviews by composing optimized search strings, screening titles and abstracts, and identifying full-text discrepancies. Third, AI assistants can draft reproducible analytic code, perform preliminary descriptive or inferential analyses, and create publication-ready tables and figures. Fourth, the models support scholarly writing by suggesting journal-specific headings, refining prose, harmonizing references, and translating technical content for multidisciplinary audiences. Fifth, they augment peer-review and editorial workflows by delivering evidence-focused critiques. In educational settings, these models can create adaptive curricula and interactive simulations for trainees, fostering digital literacy and evidence-based practice early in professional development among clinicians. Integration into clinical decision-support pipelines is also foreseeable, warranting proactive governance. Notwithstanding these opportunities, responsible use demands vigilant oversight. Large-language models occasionally fabricate citations or misinterpret domain-specific data ("hallucinations"), potentially propagating misinformation. Outputs are highly prompt-dependent, creating a reliance on informed prompt engineering that may disadvantage less technical clinicians. Moreover, uploading protected health information or copyrighted text raises privacy, security, and intellectual property concerns. We outline best-practice recommendations: maintain human verification of all AI-generated content; cross-validate references with primary databases; employ privacy-preserving, on-premises deployments for sensitive data; document prompts for reproducibility; and disclose AI involvement transparently. In summary, generative AI offers a powerful adjunct for OBGYN scientists by accelerating topic formulation, evidence synthesis, data analysis, manuscript preparation, and peer review. When coupled with rigorous oversight and ethical safeguards, these tools can enhance productivity without compromising scientific integrity. Future studies should quantify accuracy, bias, and downstream patient impact. · Generative AI supports various research stages in OBGYN, such as hypothesis generation, systematic review assistance, data analysis, and scientific writing, demonstrating its potential to streamline research workflows and improve research efficiency.. · Generative AI has notable limitations, including the risk of generating inaccurate references ("hallucinations") and the need for careful supervision.. · Effective usage requires skill in prompt engineering, posing a challenge for those without technical expertise.. · Utilizing generative AI in sensitive fields like OBGYN raises privacy, security, and ethical concerns..
1 Dec 2025
Read appraisal →Adv Anesth
The Current Use of Artificial Intelligence in Anesthesiology
Artificial intelligence (AI) applications are increasingly being used in areas of medicine such as radiology. However, within anesthesiology, there is not widespread use of AI applications at this time, though it has the potential to have a profound impact. Many applications of AI, including machine learning, deep learning, and natural language processing, are being used to create applications to enhance patient monitoring, optimize drug administration, improve decision-making, and predict postoperative complications.
1 Dec 2025
Read appraisal →