Research Appraisals
Evidence-based critical appraisals of the latest medical research, systematically evaluated using Oxford CEBM methodology.
Showing 27 appraisals
Journal of critical care
Artificial intelligence and computerized decision support in adult intensive care: A systematic review of randomized controlled trials
BACKGROUND: Artificial intelligence (AI) and computerized clinical decision support systems (CDSS) are increasingly applied in intensive care, yet their clinical impact remains uncertain, as most studies focus on model development rather than prospective evaluation. OBJECTIVES: To identify randomized controlled trials (RCTs) evaluating AI-based or CDSS interventions intended to influence real-time decision-making in adult intensive care units (ICUs) and to assess their effects on process and patient-centered outcomes. METHODS: We conducted a systematic review of randomized controlled trials (RCTs). PubMed, Embase, Cochrane CENTRAL, ScienceDirect, and IEEE Xplore were searched from inception to November 2025. Eligible studies evaluated AI-based or CDSS interventions in adult ICU patients. Study quality was assessed using RoB 2 and CONSORT-AI criteria. RESULTS: Ten RCTs were included, enrolling approximately 100,000 adult ICU patients. Five trials evaluated AI-based interventions and five evaluated CDSS. Eight trials showed improvements in at least one process measure, including earlier recognition of deterioration, improved protocol adherence, and physiological stability. However, patient-centered benefits were uncommon. Two trials reported reductions in mortality (a sepsis prediction model and a machine-learning-based early-warning system), while most studies showed no consistent effects on clinical outcomes. Reporting of AI-specific elements-dataset provenance, algorithm versioning, and human-AI interaction-was frequently incomplete. CONCLUSIONS: AI and CDSS interventions in adult ICUs are associated with improvements in process-related outcomes but show limited and inconsistent effects on patient-centered endpoints. These findings highlight a persistent gap between algorithmic innovation and clinical validation and underscore the need for pragmatic randomized trials with improved reporting and integration into clinical workflows.
3 Aug 2026
Read appraisal →Simulation in healthcare : journal of the Society for Simulation in Healthcare
Future of Biometric Technology in Healthcare Simulation
Biometric technology shows substantial promise to enhance patient safety and transform provider training. In patient care, biometric technology enables outpatient providers to optimize chronic disease by alerting providers to changes in patient conditions with real-time data. In provider training, biometric technology can inform training environments and curriculum. Certainly, there are hurdles to technology implementation including hardware fragility, cost, and software integration. Despite challenges, biometric technology has been effectively implemented in simulations and real patient settings. Looking forward, future developments focused on data security and privacy, standardizing data collection, and improving system compatibility will facilitate the wider adoption of biometric technology and improve health care quality, especially in resource-limited settings. Collaborative efforts among informatics experts, patient safety advocates, and health care providers are essential to overcome challenges and fully leverage biometric technology's potential in enhancing patient outcomes. The purpose of this commentary is to review current applications, identify key challenges, and offer recommendations to the simulation community for testing biometric technology, use in training programs, and facilitating systems integration to improve outcomes.
2 Aug 2026
Read appraisal →Rheumatic diseases clinics of North America
Demystifying Artificial Intelligence: Key Concepts with Examples in Rheumatology
Artificial intelligence (AI) refers to a broad class of computational methods to perform tasks that typically require human intelligence, such as learning patterns, reasoning, and problem solving. AI is increasingly applied across rheumatology research and clinical practice to analyze complex data and support clinical decision-making, yet its underlying concepts are often perceived as opaque or inaccessible. This aim of this article is to demystify foundational AI concepts by defining core terminology and describing major learning paradigms and analytical methods, highlighted through clinically relevant examples from rheumatology. We explain how modern AI models differ from traditional analytical approaches.
2 Aug 2026
Read appraisal →Medical physics
Reference-Free large language model agents for physician-guided radiotherapy treatment planning
BACKGROUND: Large language models (LLMs) have recently demonstrated exceptional capabilities, offering the potential to streamline workflows and enhance efficiency across diverse tasks. However, their application in domain-specific areas, such as radiotherapy treatment planning, remains challenging due to the lack of publicly available, specialized knowledge required for these tasks. PURPOSE: To investigate the critical architectural and functional components necessary to adapt an off-the-shelf large language model into a clinically viable planning agent for inverse treatment planning in intensity-modulated radiation therapy (IMRT). MATERIALS/METHODS: Twenty head-and-neck (HN) cancer patients who received IMRT at our institution were retrospectively collected under IRB approval. The LLM agent was implemented to directly interact with the clinical treatment planning system (TPS) to iteratively extract intermediate plan states and propose new constraint values to guide inverse optimization. Its decision-making was informed by real-time plan evaluations and prior optimization outcomes, enabling adaptive refinement of planning strategies across iterations. The agent was equipped with three core capabilities: comprehension of clinical planning objectives, contextual understanding of the optimization environment, and arithmetic proficiency for quantitative reasoning. Treatment planning was conducted in a reference-free inference setting, wherein the LLM operated without prior exposure to manually generated treatment plans and without any fine-tuning or task-specific training. LLM-generated treatment plans were compared with clinically approved plans created by certified dosimetrists. Key dosimetric endpoints were evaluated and statistically analyzed. Paired comparisons between LLM-generated and clinical plans were performed using the Wilcoxon signed-rank test. RESULTS: LLM-generated plans achieved comparable organ-at-risk (OAR) sparing relative to clinical plans, while demonstrating improved hot spot control (Dmax: 106.5% vs. 108.8%, p < 0.05) and superior conformity (conformity index: 1.18 vs. 1.39, p < 0.05 for boost PTV; 1.82 vs. 1.88, p = 0.47 for primary PTV). CONCLUSIONS: This study demonstrates the feasibility of a reference-free, LLM-driven workflow for automated IMRT treatment planning in a commercial TPS. The proposed approach provides a generalizable solution that could reduce planning variability and support broader adoption of AI-based planning strategies.
2 Aug 2026
Read appraisal →Cognitive, affective & behavioral neuroscience
A review of current capabilities and future directions in machine-based emotion recognition
The ability of machines to recognize emotions automatically is becoming increasingly significant across many domains where emotional understanding is essential. Such technology is applied in customer interaction, marketing, healthcare, education, the automotive industry, entertainment, and security. Providing real-time insights into human affective states improves user engagement and enables systems to respond more intelligently. Nevertheless, progress in this field is hindered by the inherent complexity of emotions, cultural differences in expression, and technical limitations that make accurate detection challenging. This paper delivers a broad review of contemporary approaches to emotion recognition. It highlights techniques based on facial expression analysis (FER), oculometrics (OM), microexpressions identification (MER), and speech analysis (SER). Further attention is given to methods involving body posture, gesture, and gait, as well as tactile interaction, text-based emotion recognition, and methods based on self-reporting. In addition, physiological signal-driven methods are discussed in depth, including respiration signals (RS), galvanic skin response (GSR), electroencephalography (EEG), electromyography (EMG), skin temperature (SKT), cardiac signals (ECG, PPG, HRV), and touch dynamics (TD) analysis. This comprehensive overview lays the foundation for advancing research on machine-based emotion recognition.
2 Aug 2026
Read appraisal →Indian journal of pediatrics
Emerging Applications of Artificial Intelligence in Pediatric Care
Artificial intelligence (AI) technologies such as machine learning (ML), deep learning (DL), predictive analytics and other tools are rapidly changing pediatric health care, using large amounts of health data. AI tools aid in triage, real-time monitoring and risk stratification in acute care settings, towards improving overall outcomes and fewer complications. During newborn resuscitation, AI analyses real-time data, can guide decisions and enhance training. Computer vision systems with AI tools can generate reliable neonatal bilirubin estimates without the need for blood sampling. AI technology is also being used in the management of necrotising enterocolitis, respiratory distress syndrome, and screening and early diagnosis of retinopathy of prematurity. ML models assist in detecting brain injuries on MRI for conditions such as hypoxic-ischemic encephalopathy, intraventricular hemorrhage; MRI biomarkers can be analyzed using AI to predict neurodevelopmental outcomes. AI-based clinical decision support systems have been deployed to enhance workflows and outcomes by early detection of disease, reducing medication errors and help clinicians improve decision-making. However, there remain ethical and practical challenges in the use of AI including data privacy, the need for high-quality pediatric datasets, rigorous clinical validation and transparency, to ensure that AI strengthens clinical judgement and is trustworthy.
2 Aug 2026
Read appraisal →Cancer medicine
Applications of Artificial Intelligence in Cancer Diagnosis and Treatment
Driven by changes in lifestyle and environmental factors, the global incidence of cancer is steadily increasing, which has established it as a leading cause of mortality worldwide. The current paradigm for cancer diagnosis and treatment relies on conventional methods, such as imaging, endoscopy, and tissue biopsy, which present significant limitations regarding sensitivity in early screening, diagnostic specificity, and personalized treatment. Consequently, the development of more efficient and accurate technologies remains a major objective in modern oncology research, and artificial intelligence (AI) has emerged as a particularly promising solution. Through machine learning and deep learning algorithms, AI is reshaping cancer care by enabling automated detection of minute lesions during screening and quantitative analysis of pathological features for diagnosis. It may also advance tumor theranostics through multimodal data integration for treatment stratification, response prediction, and image-guided or targeted therapeutic decision-making, whereas providing data-driven recommendations for personalized treatment. Despite these prospects, medical AI development faces several key issues, including data bias, model explainability, clinical reliability and generalizability, emerging limitations of foundation models and generative AI, and regulatory and ethical issues that need to be addressed. By reviewing recent advances in AI across screening, diagnosis, theranostics, and treatment, we aim to clarify where these methods are already useful, where evidence remains limited, and why closer collaboration among clinicians, engineers, and data scientists is needed for clinical translation. We hope this review serves as a practical reference for researchers and clinicians evaluating how AI may be integrated into oncology in a more standardized, clinically responsible way.
1 Aug 2026
Read appraisal →International journal of medical informatics
Machine learning models for predicting readmission after stroke: A systematic review and meta-analysis
BACKGROUND: Hospital readmission following stroke poses a significant challenge for healthcare systems. Machine learning (ML) offers the potential to improve prediction models for readmission risk, surpassing traditional statistical methods. However, the performance of ML models in such context has not been systematically evaluated. We aim to evaluate the performance of ML models in predicting post-stroke hospital readmission, as well as identifying the most important readmission predictors. METHODS: A comprehensive systematic literature review and meta-analysis was conducted.Studies included were those utilizing ML models for stroke readmission prediction. The primary outcome was the predictive performance of ML models for all time stroke readmission as reported by the Area Under the Receiver Operating Characteristic Curve (AUROC). Pooled AUROC values were calculated using a random-effects model. RESULTS: A total of eleven studies involving 380,254 patients were analyzed. Mean age of the patients ranged between 64.5 and 79.7 years. In total, 49 ML models were reported with Logistic regression (LR) and Random Forest being the most used. The number of input predictive variables used for model training ranged from 6 to 10,047. The overall pooled AUROC for predicting hospital readmission was 0.74 (95% CI: 0.69 to 0.78). For 30-day readmission, the pooled AUROC was 0.73 (95% CI: 0.67 to 0.79) while for 90-day readmission, it was 0.75 (95% CI: 0.69 to 0.80). The most frequently reported variables as being top predictors of readmission were Length of stay (LOS) (3 studies), Age (2 studies), National Institutes of Health Stroke Score (NIHSS) (2 studies), HbA1c (2 studies), Homocysteine blood level (2 studies). CONCLUSION: Machine learning models demonstrate moderate predictive performance in stroke readmission risk prediction. Future research should focus on validating and refining existing models and adopting unified methodological approaches to aid in drawing more accurate conclusions.
1 Aug 2026
Read appraisal →Balkan medical journal
Bias and Fairness Across the Healthcare AI Lifecycle: A Clinician-Oriented Review
Artificial intelligence (AI) is increasingly being investigated and, in selected clinical settings, implemented to support diagnosis, triage, and workflow optimization. Although these systems have the potential to improve access, consistency, and efficiency, they may also reproduce or amplify health inequities when bias is introduced during development, evaluation, implementation, or postdeployment use. This clinician-oriented narrative review adopts a practical lifecycle approach to explain how algorithmic unfairness becomes clinically relevant, how clinicians can recognize it, and how institutions can mitigate its impact. We first outline the ethical, clinical, and mathematical dimensions of fairness. We then examine fairness risks and sources of bias across six stages of the healthcare AI lifecycle: problem formulation, data generation, model development, evaluation, implementation, and postdeployment monitoring and governance. Key mechanisms include biased proxy outcomes, unrepresentative or error-prone data and labels, model shortcut learning, hidden stratification, distribution shift, and human-AI interaction effects (e.g., automation bias and alert fatigue), all of which can create feedback loops and contribute to fairness drift over time. For each stage, we identify clinician-facing red flags and practical mitigation strategies, including defining clinically meaningful outcomes, using representative and well-documented datasets, conducting subgroup-stratified evaluations, performing external and prospective validation, justifying decision thresholds, implementing safeguards for human-AI interactions, and maintaining continuous postdeployment monitoring, including postmarket surveillance for regulated medical devices. Fairness cannot be ensured through a single metric, publication, regulatory clearance, or one-time validation. Instead, equitable healthcare AI requires transparent design, rigorous evaluation, local governance, and ongoing monitoring across diverse populations, clinical sites, devices, workflows, and time. Fairness should therefore be regarded as a continuous clinical and institutional responsibility rather than a downstream technical consideration.
1 Aug 2026
Read appraisal →Scientific reports
Manual federated simulation for multiple sclerosis integrating XGBoost algorithm with SHAP explanation
Multiple sclerosis (MS) is a chronic autoimmune disorder of the central nervous system, underscoring the importance of early and accurate diagnosis. In this study investigates the predictive modelling of MS progression in patients with Clinically Isolated Syndrome (CIS), privacy-preserving for a federated and explainable Machine Learning (ML) framework. To address missing data while preserving inter-feature dependencies, Multivariate Imputation by Chained Equations (MICE) with iterative imputers was employed. Classification was performed using the Extreme Gradient Boosting (XGBoost) algorithm. Model interpretability was developed through Explainable Artificial Intelligence (XAI) techniques, specifically Shapley Additive Explanations (SHAP). To ensure data confidentiality and simulate decentralized clinical environments, an in silico federated learning framework was applied. Experimental results demonstrated strong predictive performance, achieving 96.7% accuracy and 99% ROC-AUC during training, 92.5% accuracy in validation, and 81.8% accuracy with an AUC of 88% on the test set. For the Federated Learning (FL) simulation, the model maintained competitive performance, yielding an accuracy of 76.3% and an AUC of 83.9%. The proposed approach supports early diagnosis, enhances clinical trust through interpretability, and promotes secure data collaboration, thereby contributing to more informed and transparent clinical decision-making and improved patient care.
1 Aug 2026
Read appraisal →Journal of medical Internet research
Shadow AI in Swedish Health Care: Qualitative Analysis of Physicians' Free-Text Answers
BACKGROUND: The rapid emergence of artificial intelligence (AI) has outpaced its formal adoption in health care organizations, contributing to the emergence of Shadow AI, defined here as the use of unauthorized AI tools by medical professionals. Under the European Union Medical Device Regulation, AI tools used for clinical purposes must undergo conformity assessment before use; general-purpose tools such as ChatGPT have not done so, rendering their clinical application unauthorized at the regulatory level. While Shadow AI offers potential efficiency gains and higher performance, it poses significant risks to data privacy, clinical safety, and regulatory compliance. Despite its growing prevalence, empirical research on the purposes for which physicians use Shadow AI remains scarce. OBJECTIVE: This study explores the purposes for which physicians describe using Shadow AI in their work. METHODS: We conducted a cross-sectional survey of physicians employed in Swedish health care organizations (N=357; response rate~64%). Data were collected between December 2023 and January 2024 via a verified online panel. We conducted a qualitative content analysis of free-text responses on the use of unauthorized AI tools. We applied theoretical lenses from the sociology of professions and paradox theory to interpret the empirical findings. RESULTS: Physicians use Shadow AI for several purposes, which we grouped into 4 categories: clinical work and decision-making, administrative work, research and professional development, and technological interest and curiosity. More specifically, Shadow AI is used as a colleague and second opinion for clinical decision support (eg, differential diagnoses and rare cases), administrative tasks such as patient communication and documentation, and research aimed at staying up to date and exploring developments in generative AI. Physicians described using these tools compensated for perceived gaps in institutional systems, reducing workload, and accessing knowledge considered difficult to obtain through conventional channels. The findings reveal a tension between physicians' drive to improve their practice and the regulatory and organizational constraints that render such use unauthorized. CONCLUSIONS: Shadow AI used by physicians presents both opportunities and risks for health care professionals and organizations. Shadow AI indicates gaps where formal hospital systems may fail to meet health care professionals' needs and signals a way for physicians to strengthen their experience-based knowledge. It represents a renegotiation of professional boundaries, as physicians bypass institutional constraints to maintain professional efficacy. The findings highlight a paradox in which the same tools that pose regulatory and safety risks also address real gaps in clinical and administrative support, suggesting that governance approaches must account for this tension rather than relying on prohibition alone.
30 July 2026
Read appraisal →Journal of medical Internet research
Initial-Visit Specialty Triage in Rare Diseases Using Large Language Models: Retrospective Benchmarking Study
BACKGROUND: Specialty triage at first contact is an overlooked step in early diagnostic pathways for rare diseases. Patients often present with overlapping, multisystem, and atypical manifestations, making first-visit specialty selection challenging and potentially prolonging diagnostic pathways. OBJECTIVE: The aim of this study is to evaluate the accuracy, response time, and consistency of large language models (LLMs) for initial-visit specialty triage in rare diseases across multiple datasets, and to compare their performance with registered nurses and nonmedical participants. METHODS: In this retrospective benchmarking study, we used 5 rare disease datasets: a publication-derived case set, 3 RareBench-derived datasets, and a Facial phenotype-Gene-Disease Dataset-derived set. Fourteen LLMs were evaluated over 5 independent runs per case. Performance was assessed using accuracy, response time, and consistency, with subgroup analyses by model accessibility, reasoning mode, parameter scale, and phenotype count. Human comparison was conducted on the publication-derived case set using registered nurses and nonmedical participants. RESULTS: Across datasets, model accuracy ranged from 0.4378 to 0.7141. Claude-opus-4-5 achieved the highest accuracy (0.7141) and consistency (0.9653), averaging 10.79 seconds per case. GPT-5.1 had the shortest response time (3.39 s/case) and high accuracy (0.6948). Proprietary models had numerically higher average accuracy than open-weight models (0.6973 vs 0.6365). Nonthinking models achieved higher average accuracy than thinking models (0.6789 vs 0.5826) and had shorter response times, although this exploratory comparison was based on a small number of thinking models. Accuracy varied by phenotype count, with higher performance in cases with 1 to 2 or more than 14 phenotypes. On the publication-derived case set, LLMs achieved higher average accuracy than registered nurses and nonmedical participants (0.5978 vs 0.4914 and 0.4573). CONCLUSIONS: LLMs showed potential as assistive tools for initial-visit specialty triage in rare diseases. Model choice, reasoning mode, and phenotype information density influenced performance, but subgroup findings should be interpreted cautiously. Future work should evaluate LLM-based specialty triage in prospective clinical settings and develop clinician-supervised workflows with traceable evidence support.
24 July 2026
Read appraisal →Journal of medical Internet research
Model and Task-Aware Test-Time Scaling Strategies for Large Language and Vision-Language Models in Medicine: Evaluation Study
BACKGROUND: Test-time scaling has emerged as a promising method to enhance the reasoning capabilities of large language models (LLMs) and vision-language models (VLMs) during inference without additional training. While foundational studies established scaling paradigms in general domains, their applicability to the unique complexities of medical AI remains underexplored. OBJECTIVE: This study aims to conduct a comprehensive investigation of test-time scaling in the medical domain. We evaluate the impact of scaling across different model sizes and task complexities. Furthermore, we seek to identify domain-specific bottlenecks and assess model robustness against user-driven perturbations, such as misleading clinical authority. METHODS: This study evaluated a diverse set of general and medical-specific LLMs and VLMs. Experiments used five textual medical benchmarks comprising over 5500 questions and two multimodal benchmarks comprising 7000 samples. Performance was measured under three scaling conditions: increasing token budgets, iterative sequential scaling, and parallel scaling. Robustness was tested by embedding misleading hints with varying tones and levels of simulated clinical expertise into prompts. RESULTS: For nonreasoning LLMs, accuracy saturated quickly, with token usage often remaining under 500 tokens regardless of budget increases. Reasoning models demonstrated significant performance gains on complex tasks as token budgets increased. Notably, we identified distinct domain-specific behaviors. First, current VLMs showed a structural bottleneck in integrating visual clues and experienced limited benefit from token expansion. Second, medically fine-tuned LLMs excelled in clinical question answering but exhibited degraded scaling efficiency on calculation tasks compared to general-domain models. This reflects a disparity between qualitative clinical alignment and procedural logic. Third, while optimal scaling improved robustness, models exhibited a cognitive vulnerability by readily abandoning correct reasoning when confronted with misleading expert physician hints. Regarding scaling strategies, parallel scaling outperformed sequential scaling on easier tasks. Conversely, extended sequential scaling or increased budgets proved essential for complex problem-solving. CONCLUSIONS: Test-time scaling rules from general domains do not perfectly translate to medical AI. Longer reasoning is not universally beneficial. Concise reasoning with parallel scaling is optimal for simpler tasks. An extended chain of thought via sequential scaling or increased budgets is required for complex problems. Furthermore, safe clinical deployment requires addressing fundamental vision-language alignment, balancing clinical and procedural reasoning, and mitigating vulnerabilities to perceived clinical authority.
24 July 2026
Read appraisal →Journal of medical Internet research
Evaluation Frameworks for Clinical AI Incorporating Validation Strategies, Real-World Applicability, and Ethical Principles: Scoping Review
BACKGROUND: AI shows substantial potential in health care; however, the absence of standardized evaluation frameworks limits its safe and effective clinical implementation because of inconsistent validation requirements and fragmented ethical principles. Existing guidelines vary in structure, methodological rigor, and ethical integration, creating uncertainty. OBJECTIVE: This study aimed to systematically map, characterize, and critically analyze existing evaluation frameworks for clinical AI, focusing on three core dimensions: methodological rigor, validation strategies (internal validation, including reporting of technical and clinical performance; external validation, including real-world applicability), and alignment with the United Nations Educational, Scientific and Cultural Organization (UNESCO) AI ethical considerations. METHODS: A scoping review was conducted following PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines. Six databases (PubMed, Embase, BVS, EBSCOhost, ProQuest, and Sage) and the Enhancing the Quality and Transparency of Health Research Network were searched without language or date restrictions up to February 2026. Eligible documents included peer-reviewed papers, gray literature, and organizational guidelines describing evaluation or reporting frameworks for clinical AI. Editorials, commentaries, and conference abstracts lacking a clearly defined evaluative framework or clinical applicability were excluded. Two reviewers independently screened records and extracted data. Data were extracted across three domains: (1) general characteristics, (2) methodological rigor and validation parameters, and (3) ethical integration and were synthesized using a dot plot-based gap map. Ethical adherence was assessed using a 10-domain UNESCO-based scoring matrix. No formal risk-of-bias assessment was conducted, consistent with scoping review methodology. RESULTS: From 3363 records, 46 frameworks met the inclusion criteria. Mapping revealed a rapidly expanding but fragmented landscape. Most frameworks targeted investigational use (88%), with limited focus on clinical applicability. Frameworks varied in structure, methodology, and scope, with a predominance of reporting guidelines and few validated tools. Most (63%) were developed through multi-institutional collaborations, and 32.6% incorporated transdisciplinary participation. Only 31.8% reported technical metrics (commonly area under the curve, sensitivity, and specificity), and 15.9% provided clinical indicators (eg, predictive values or calibration). Only 11.4% achieved methodological rigor, incorporating validation aligned with intended use, while most relied on partial validation strategies, highlighting a gap between model development and clinical evaluation. Ethical integration was heterogeneous: only 5 frameworks achieved high compliance (≥80%), whereas 4 scored <10%. The most frequently addressed UNESCO principles were awareness and education (71.1%) and transparency and explainability (70%), while human oversight (24.4%) and adaptive governance (33.3%) were least represented. Findings indicate a misalignment between framework design, validation requirements, and clinical implementation. CONCLUSIONS: Evaluation frameworks for clinical AI remain heterogeneous and oriented toward investigational contexts. Critical gaps persist in methodological rigor, validation aligned with intended use, and fragmented ethical coverage. These findings highlight the need for standardized, robust, and ethically grounded frameworks to enable safe, reliable, and scalable integration of AI into clinical practice.
24 July 2026
Read appraisal →PloS one
A novel dynamic Optuna hybrid Harris Hawks Optimization approach for classification of CAD
Coronary Artery Disease (CAD) is a leading cause of mortality worldwide and is primarily associated with atherosclerotic plaque formation, resulting in coronary artery stenosis. The accurate prediction of CAD using deep learning models is often constrained by the limitations of conventional optimization techniques, including premature convergence and limited adaptability of the models. To address these challenges, this study proposes a Dynamic Optuna Hybrid Harris Hawks Optimization (Dynamic Optuna H-HHO) framework to enhance the performance of deep learning-based CAD prediction models. The proposed approach integrates dynamic parameter adjustment, adaptive escape energy mechanisms, and Optuna-based hyperparameter tuning. The framework was applied to optimize several deep-learning classifiers, including ResNet-50, VGG-16, InceptionV3, and MobileNet, using a coronary artery stenosis dataset. The performance was evaluated through a comparative analysis with models optimized using the conventional Hybrid Harris Hawks Optimization (H-HHO) algorithm. The experimental results indicate that the proposed Dynamic Optuna H-HHO framework consistently improves the predictive accuracy across all evaluated models. InceptionV3 achieved the highest accuracy of 97.9%, followed by MobileNet with 97.6%, compared with the maximum accuracy of 82.46% obtained using traditional HHO-based optimization. By combining adaptive optimization strategies with automated hyperparameter tuning, the proposed framework provides a robust and scalable solution for improving the accuracy of coronary artery disease prediction.
20 July 2026
Read appraisal →Scientific reports
Walking stability prediction for pedestrians using gait energy images and hybrid deep and few-shot learning models
The prediction and recognition of unstable human walking patterns are of high importance for active video surveillance, smart environments, and assistive healthcare, particularly for fall detection in the elderly. This research investigates the utility of Gait Energy Images (GEIs) combined with deep learning, vision transformers, and few-shot learning architectures to enhance the classification of stable and unstable pedestrian walking patterns. We evaluate and compare twelve methodologies: six classical or feature-based machine learning models (Linear SVM, HOG + SVM, LBP + RBF-SVM, Random Forest, XGBoost, and an adapted GaitSet baseline), three deep learning models (MobileNet, Vision Transformer (ViT), and YOLOv8-cls), and three episodic few-shot learning techniques (Prototypical, Matching, and Relation Networks) under data-scarcity regimes. To facilitate this evaluation, we introduce the Unstable and Stable Walking Pedestrian (USWP) dataset, constructed by fusing and harmonizing sequences from seven public action recognition databases, containing 3250 unique GEIs with a subject-independent evaluation protocol to prevent identity-based domain leakage. Our experiments demonstrate that the YOLOv8-cls model achieves an overall accuracy of 96.92% (97.14% on Loss-of-Balance anomalies and 94.67% on Active Motion anomalies), significantly outperforming the conventional Linear SVM baseline (75.38%) and MobileNet (91.08%). Conversely, Relation Networks exhibit lower few-shot performance (71.08%) due to optimization complexities in learning similarity metrics from sparse data. Leave-One-Dataset-Out (LODO) cross-validation reveals an average generalization accuracy of 83.28%, indicating significant domain bias across source databases and underscoring that within-dataset evaluations overestimate real-world generalization. Computational complexity analysis shows that MobileNet provides an optimal trade-off for real-time edge deployment (4.2 ms latency), while preprocessing ablation studies demonstrate that integrating the Segment Anything Model (SAM) with MediaPipe-derived Regions of Interest (ROIs) yields a 12.50% absolute improvement in accuracy by eliminating background noise.
19 July 2026
Read appraisal →Sleep
AI meets sleep surgery: assessing drug-induced sleep endoscopy interpretation with a large language model.
STUDY OBJECTIVES: To evaluate the performance and safety of a large language model in interpreting drug-induced sleep endoscopy (DISE) videos and providing treatment recommendations for obstructive sleep apnea, compared to expert human raters and the contemporaneous clinical report. METHODS: This prospective, blinded study included 16 adults undergoing drug-induced sleep endoscopy at a tertiary academic center. For each case, an anonymized procedural video, clinical vignette, and examination findings were independently reviewed by two sleep surgery experts, a senior otolaryngology resident, and a large language model. All the raters assessed the video quality, airway maneuvers, airway collapse at each anatomical site, and recommended therapy. Concordance with the clinical reference standard was evaluated using Cohen's kappa and intraclass correlation coefficients. Safety and reproducibility were assessed through subgroup and error-type analyses. RESULTS: Human raters demonstrated perfect or near-perfect agreement with the reference standard for the velum, oropharynx, tongue base, and jaw thrust response, and moderate agreement for epiglottic collapse. The large language model matched expert performance for all domains except the epiglottis, where moderate agreement was observed. Model-generated treatment recommendations were safe, consistent with accepted clinical practice, and highly reproducible between independent runs. No unsafe or discordant recommendations were identified. CONCLUSIONS: In this single-center study of 16 patients, a large language model accurately interpreted DISE videos and generated safe recommendations consistent with accepted clinical practice, approximating expert performance in most domains. Larger multicenter cohorts are needed to validate these findings and confirm generalizability. Statement of Significance This study demonstrates that artificial intelligence can interpret complex airway videos in sleep surgery and provide safe, expert-level treatment advice. By comparing the performance of a large language model with that of experienced clinicians, our findings suggest that advanced technology can help standardize decision-making in a highly subjective area of care. This work highlights the promise of artificial intelligence as an adjunct to clinical judgment, especially in settings where expert access is limited. However, important questions remain about the use of artificial intelligence in challenging cases and its integration into real-world practice. Future research should focus on larger, more diverse groups of patients and on ensuring ongoing oversight and safety.
14 July 2026
Read appraisal →JMIR research protocols
Applications of Large Language Models in Ovarian Cancer Management: Protocol for a Systematic Review and Meta-Analysis
BACKGROUND: Ovarian cancer (OC) is a highly fatal gynecologic malignancy with complex management challenges and limited long-term survival for advanced stages. Large language models (LLMs)-including systems such as GPT-4, Claude, Google Gemini, and others-are emerging artificial intelligence (AI) tools capable of performing health care-related tasks such as diagnostic support, treatment planning, report generation, and patient communication. However, their applications in OC care have not yet been comprehensively assessed. OBJECTIVE: This protocol outlines a systematic review and meta-analysis aimed at evaluating the use, performance, and clinical impact of LLMs in OC management. We will examine how LLMs have been applied across various domains (eg, diagnosis, prognosis, treatment planning, and patient engagement), the metrics used to assess their performance (eg, accuracy, sensitivity, and area under the curve), and their strengths and limitations. METHODS: This review will be conducted in accordance with PRISMA-P (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Protocols) guidelines. A comprehensive search strategy will be implemented across biomedical, technical, and Chinese-language databases (eg, PubMed, Embase, Web of Science, IEEE Xplore, and China National Knowledge Infrastructure) from inception to December 31, 2025. Eligible studies include clinical evaluations, validation studies, and real-world implementation reports involving LLMs in OC care. Two independent reviewers will perform screening, data extraction, and quality appraisal using validated tools (eg, version 2 of the Cochrane risk-of-bias tool for randomized trials, Risk of Bias in Nonrandomized Studies of Interventions, Quality Assessment of Diagnostic Accuracy Studies 2, and Prediction Model Study Risk of Bias Assessment Tool+AI). Outcomes of interest include model performance metrics, clinical process impacts, safety concerns, and usability. Meta-analyses will be conducted where feasible using random-effects models in R (meta, metafor, and mada packages), including bivariate models for sensitivity and specificity. RESULTS: The review is currently in progress. The PROSPERO registration has been completed, and the literature search and selection process is underway. Study selection, data extraction, and quality assessment are expected to be completed by mid-2026. Final results will include pooled performance metrics (eg, accuracy, F1-score, and area under the curve), qualitative insights into clinical integration, and identification of limitations such as reporting bias or insufficient external validation. CONCLUSIONS: This systematic review will provide the first comprehensive synthesis of evidence on the application of LLMs in OC care. It will identify promising use cases, highlight safety and reporting challenges, and inform future research directions. The findings are expected to support evidence-based integration of LLMs into gynecologic oncology workflows while promoting transparency and methodological rigor in AI evaluation.
12 July 2026
Read appraisal →Journal of medical Internet research
Attitudes and Needs of Health Care Providers Toward Artificial Intelligence-Assisted Pediatric Palliative Care: Mixed Methods Study
BACKGROUND: While artificial intelligence's (AI's) transformative potential in health care is widely acknowledged, its application in highly sensitive, humanistic domains like pediatric palliative care (PPC) remains largely unexplored. OBJECTIVE: This study aims to explore the attitudes and needs of health care providers on the PPC assisted by AI, with the goal of informing future development and implementation of AI systems in this field. METHODS: This was an explanatory sequential mixed methods study consisting of a nationwide cross-sectional questionnaire survey (March-April 2025) followed by qualitative semistructured interviews (August-October 2025). The quantitative study aimed to investigate PPC health care providers' experiences, attitudes, and needs for the application of AI. Participants included team members of all recognized PPC teams in mainland China. The qualitative study aimed to explore in greater depth the potential future roles of AI in this field, as well as the features of an ideal AI-assisted tool for PPC. Potential interviewees were recruited from the pool of quantitative survey respondents. RESULTS: Among 352 survey respondents, most (n=205, 58.24%) reported moderate familiarity with AI, with large language models being the most commonly used (n=280, 79.55%). Among large language model users, over half (161/280, 57.50%) reported using them for clinical purposes. Attitudes were generally positive: 67.05% (236/352) believed AI's benefits would outweigh drawbacks, and 75% (264/352) considered its implementation feasible. The most desired applications were patient and family education (276/352, 78.41%) and symptom management (257/352, 73.01%). Interviews with 17 providers revealed three themes: (1) clinical roles and boundaries, (2) elements for clinical integration, and (3) challenges in development and deployment. CONCLUSIONS: This study reveals that PPC providers express positive attitudes and strong demand for AI-assisted clinical work. Furthermore, the research clarifies appropriate roles for AI, outlines elements for clinical integration, and highlights potential challenges in development and integration. This study provides evidence for the feasibility of AI application in PPC and offers guidance for the future development and deployment of AI tools.
5 July 2026
Read appraisal →PloS one
Deep learning based two-way feature depiction model for brain tumor detection
Brain tumors are one of the most fatal disorders that cause one of the highest mortalities in the world. Gliomas are the most common primary brain tumors originating from glial cells in the central nervous system. Traditionally, a tissue sample is extracted and examined for its genetic and characteristic properties. This method is invasive, painful, and takes a longer period to produce results. Various automatic Deep learning (DL) based schemes have been presented for the brain glioma detection, but they lack due to poor explainability, lower generalization, poor feature depiction, class imbalance problem and lower detection rate. This paper presents a deep learning based brain tumor detection using two way feature depiction model (TWFDM) that combines the 2D-Deep Convolution Neural Network (DCNN) and 1D-DCNN. The 2D-DCNN accepts the raw MRI images and the 1D-DCNN accepts the handcrafted local binary pattern (LBP), gray level cooccurrence matrix (GLCM), and Histogram of Oriented Gradient (HOG) features. Furthermore, improved particle swarm optimization (IPSO) is used for feature selection to minimize the computational complexity of the TWFDM system. The proposed TWFDM achieves an overall accuracy of 96.25%, a recall of 96.34%, a precision of 96.31%, and an F1-score of 96.32% on the Brain MRI dataset for four-class classification, representing an important improvement over traditional techniques.
3 July 2026
Read appraisal →Journal of the American Medical Informatics Association : JAMIA
TRUST: an large language model-based dialogue system for trauma understanding and structured assessments
OBJECTIVES: While large language models (LLMs) have been widely used to assist clinicians and support patients, no existing work has explored dialogue systems for standard diagnostic interviews and assessments. This study aims to bridge the gap in mental healthcare accessibility by developing an LLM-powered dialogue system that replicates clinician behavior. MATERIALS AND METHODS: We introduce TRUST, a framework of cooperative LLM modules capable of conducting formal diagnostic interviews and assessments for post-traumatic stress disorder (PTSD) following the Clinician-Administered PTSD Scale for DSM-5 (CAPS-5). To guide the generation of appropriate clinical responses, we propose a Dialogue Acts schema specifically designed for clinical interviews. Additionally, we develop a patient simulation approach based on real-life interview transcripts to replace time-consuming and costly manual testing by clinicians. RESULTS: A comprehensive set of evaluation metrics is designed to assess the dialogue system from both the agent and patient simulation perspectives. Expert evaluations by conversation and clinical specialists show that TRUST performs comparably to real-life clinical interviews. DISCUSSION: Our system performs with clinical quality approaching that of human clinicians, with room for future enhancements in communication styles and response appropriateness. CONCLUSIONS: Our TRUST framework shows its potential to facilitate mental healthcare availability.
2 July 2026
Read appraisal →Joint Commission journal on quality and patient safety
Large Language Models as Physician Recommenders: Limitations, Alternatives, and the Path Toward Hybrid AI Systems
Digital tools are increasingly used by patients to access health information and navigate care, including choosing clinicians, but the evidence supporting "best doctor" recommendations varies widely across available methods. This manuscript compares the emerging use of consumer-facing large language models (LLMs) for this purpose with four alternative physician-selection strategies: patient networks, expert referrals, consumer rating/search platforms, and outcome-informed artificial intelligence / machine learning (AI/ML) recommender systems. Consumer rating platforms mainly reflect patient experiences rather than technical skills and often show weak or inconsistent links with objective performance measures. General-purpose LLMs may produce convincing rationales that highlight visibility and reputation, but they remain limited by hallucinations, prompt sensitivity, and uncertainty about whether their explanations truly reflect the underlying reasoning. Outcome-informed AI/ML recommender systems could, in theory, incorporate risk-adjusted clinical performance, but they face significant challenges, including unreliable physician-level measurement, incomplete data, residual confounding, and limited transparency, which can undermine trust and informed consent. We suggest that a hybrid system should be seen as a guiding concept and a testable approach rather than a final solution. Such a system could integrate proven, auditable performance measures with transparent patient-preference filters and verifiable, patient-facing explanations. We present key design principles, including clinical relevance, explainability, autonomy, equity, and governance, in line with current US Food and Drug Administration (FDA) guidance related to clinical decision support and adaptive AI life cycle management.
1 July 2026
Read appraisal →The British journal of radiology
Artificial intelligence in nuclear medicine
Artificial intelligence (AI) holds great promise for advancing diagnostics and treatment in nuclear medicine. The rapid growth of AI over the past decade has been largely driven by advances in hardware components such as graphics processing units (GPUs) and the introduction of deep learning (DL) and convolutional neural networks (CNN). The integration of AI and medical imaging has the potential to revolutionize nuclear medicine by, for example, accelerating image acquisition, enhancing image quality, enabling advanced image generation, assisting image interpretation, and aiding treatment planning. Clinical applications have been demonstrated for most medical specialties, including oncology, neurology, and radionuclide therapy. The utilization of AI to provide automated, standardized procedures can help bring advanced imaging from major university centres to smaller local clinics, thus benefiting a broader range of patients. Additionally, AI has vast potential for predicting optimal treatment strategies, assessing risk, optimizing patient flow and outcomes, and even improving productivity, but these capabilities have yet to be fully utilized. The fraction of clinical AI applications in general healthcare reaching beyond the prototyping phase is reported to be as low as 2%. Indeed, in nuclear medicine, very few AI developments have reached commercial maturity. Currently, most AI applications in nuclear medicine follow the imaging flow from image acquisition and reconstruction, post-processing and image preparation, image analysis, and decision support for clinical interpretation. Below, we will briefly review selected areas and comment on challenges and opportunities for AI in nuclear medicine, with a special focus on the transition from development to clinical implementation.
1 July 2026
Read appraisal →Current allergy and asthma reports
Harnessing Machine Learning and Electronic Health Record Data to Improve Asthma Management
PURPOSE OF REVIEW: The review examines the application of machine learning (ML) and large language models (LLMs) to asthma management. We sought to identify clinically relevant applications, and particularly those that harness electronic health record data. We review methodological challenges and future directions for translating these tools into meaningful improvements in asthma care. RECENT FINDINGS: ML applied to electronic health record data has been utilized across several domains of asthma management: predicting medication response to inhaled corticosteroids and biologics, improving inhaler adherence through digital inhaler systems, and predicting exacerbation risk with moderate-to-high accuracy. Tools for patient education include clinician-guided chatbots, as well as publicly available LLMs. Limitations include accuracy, hallucinations, and patient health literacy. ML and LLMs offer promising pathways towards personalized, data-driven asthma management. However, harnessing this potential will require rigorous external validation, transparent model design, equitable implementation, and adaptive clinician oversight. Collaboration among clinicians, data scientists, and policymakers will be essential to implement these tools for patient-centered asthma care.
30 June 2026
Read appraisal →PloS one
Few-shot skin lesion classification with Adaptive Multi-Scale Convolutional Attention Network
Computer-aided diagnosis of skin lesions faces core challenges, including scale diversity, blurred boundaries, intra-class morphological variations, and sparse data. Existing methods often rely on fixed receptive fields or generic attention mechanisms, struggling to fully adapt to the unique characteristics of skin lesions. To address this, we propose an Adaptive Multi-scale Convolutional Attention Network (AMCANet), which aims to achieve accurate and robust classification of skin lesions with limited data. AMCANet comprises three core modules: the adaptive multi-scale convolution module dynamically adjusts the receptive field to accommodate lesions of varying sizes; the hierarchical channel attention module integrates multi-level semantic information across different resolutions; and the skin spatial attention module leverages image gradient information to enhance lesion boundaries and local texture features. Extensive few-shot experiments on the HAM10000 and PAD-UFES-20 public datasets demonstrate that AMCANet significantly outperforms existing baseline models across multiple metrics, exhibiting promising generalization capabilities on the evaluated datasets. Qualitative and visual analyses further validate the model's ability to extract discriminative features and effectively focus on lesion regions. This study proposes a deep learning model, which demonstrates certain effectiveness in classifying skin lesions even with a small number of samples, providing a potential direction for future research.
27 June 2026
Read appraisal →Scientific reports
Multimodal emotion recognition using hybrid deep feature fusion under speaker-independent evaluation
Emotion recognition is one of the most important and complex challenges for machines to understand, as most robots and AI agents struggle with human-centric perception and interpretation. Therefore, this paper introduces a novel multimodal emotion recognition system that analyzes emotions through two complementary channels: voice and facial expressions. The proposed approach is evaluated on the RAVDESS and CREMA-D datasets, which consist of acted emotional expressions across multiple discrete emotion categories. Utilizing an advanced multimodal deep feature fusion technique, the system combines handcrafted audio features (e.g., Mel-Frequency Cepstral Coefficients (MFCCs)) with deep visual features extracted from an attention-based VGGFace model. These features are integrated into a unified representation through a hybrid fusion strategy that jointly employs concatenation, cross-attention, gated fusion, and multiplicative fusion mechanisms to capture complementary cross-modal interactions. To ensure a comprehensive and realistic assessment, the model is evaluated under both random-split and strict speaker-independent protocols. On the RAVDESS dataset, the proposed system achieves an accuracy of 95.83% under random-split evaluation and 48.06% ± 9.76% accuracy under speaker-independent Leave-One-Speaker-Out (LOSO) testing, while on the CREMA-D dataset it attains 73.54% accuracy using random splits and 53.12% ± 2.65% accuracy under subject-exclusive speaker-independent 5-fold cross-validation.
27 June 2026
Read appraisal →PloS one
Artificial Intelligence in emergency department triage: A scoping review
BACKGROUND: Triage in emergency departments (ED) is a critical process for prioritizing care and ensuring clinical safety. However, current triage systems often exhibit vulnerabilities that compromise the efficiency and quality of healthcare delivery. Artificial Intelligence (AI) has emerged as a promising innovation to support decision-making and optimize patient flow in these high-pressure environments. OBJECTIVE: To map the available evidence regarding the implementation and performance of artificial intelligence in emergency department triage. METHOD: This scoping review followed the Joanna Briggs Institute (JBI) methodology and the PRISMA-ScR guidelines. A comprehensive search was conducted across 13 databases (CINAHL, Cochrane Library, PubMed Central, SciELO, Web of Science, SCOPUS, Science Direct, VHL, Embase, and several regional dissertation repositories), with no language or time restrictions. Two independent reviewers performed the selection process using the Rayyan platform, with discrepancies resolved by a third evaluator. Data were synthesized using the PAGER framework, categorizing findings into Patterns, Advances, Gaps, Evidence for practice, and Recommendations for research. RESULTS: Nineteen studies met the inclusion criteria. AI was primarily implemented through Machine Learning (ML) algorithms, including Deep Learning architectures. Natural Language Processing (NLP) was frequently employed to process unstructured clinical data, with recent studies exploring the potential of Large Language Models (LLMs). Overall, ML-based models consistently outperformed traditional triage systems in predictive accuracy. These techniques were mainly utilized for automated classification, predicting clinical severity, and enhancing patient prioritization by integrating both objective and subjective assessment data. CONCLUSIONS: The findings indicate that AI has significant potential to enhance emergency triage by streamlining service flows and providing robust clinical decision support. However, the current evidence remains heterogeneous and largely exploratory. Key challenges include variability in model performance, a lack of external validation, and studies often limited to specific populations. Consequently, many current tools still lack the necessary reliability for safe, large-scale clinical implementation.
27 June 2026
Read appraisal →