Research Appraisals
Evidence-based critical appraisals of the latest medical research, systematically evaluated using Oxford CEBM methodology.
Showing 4 appraisals
Physiological measurement
A comprehensive inference-time augmentation framework in physiological signals: application to PPG-based AF detection
Objective.Accurate classification of physiological signals in real-world deployments is challenged by sensor noise, motion artifacts, and distribution shifts between training and deployment data. Inference-time augmentation (ITA), which applies augmentations during inference rather than retraining, offers a simple, model-agnostic mechanism to improve robustness. However, ITA application to physiological signals has remained narrow in scope, relying on limited augmentation methods with fixed, unoptimized parameters. This work proposes a unified ITA framework to address that gap.Approach.The framework incorporates 13 augmentation methods spanning time-domain, amplitude-domain, frequency-domain, and artifact-injection transformations, with hyperparameters systematically optimized via Bayesian optimization. We evaluate the framework on atrial fibrillation (AF) detection from 30 s photoplethysmography (PPG) signals using two deep learning architectures, generative pre-trained transformer (GPT)-PPG (in two sizes) and ResNet, across five datasets comprising more than 400 patients and ∼9800 h of PPG recording. Two evaluation strategies are assessed: standard ITA applied to all inputs, and selective ITA applied to initially positive predictions.Main results.Standard ITA consistently improved the area under the receiver operating characteristic curve (AUROC, up to 8.5% for GPT-PPG and 0.7% for ResNet) and the area under the precision-recall curve (AUPRC, up to 10.6% for GPT-PPG and 0.8% for ResNet) across all model-dataset combinations. Selective ITA further reduced average FPR by up to 4.4% (GPT-PPG) and 1.3% (ResNet) on non-AF PPG datasets.Significance.These findings establish ITA as a practical, model-agnostic approach for improving the reliability of PPG-based AF classification in deployment settings where retraining is not feasible, with broader applicability to physiological signal analysis.
9 July 2026
Read appraisal →Scientific reports
Cross-domain transfer learning strategy enhances interpretability of deep learning model explanations
Clinical decision-making increasingly relies on deep neural networks (DNNs), yet their deployment in practice requires transparent and interpretable predictions. Explainable artificial intelligence (xAI) methods can identify input regions relevant to a model's decision, but their clinical interpretability remains limited. In this study, we investigated whether inductive transfer learning (TL) can reinforce domain-specific feature separation in xECGArch, a two-branch convolutional neural network for atrial fibrillation (AF) detection from electrocardiograms (ECGs). Each branch was pre-trained on a task aligned with its designated feature domain, P wave detection for morphology and RR interval variability prediction for rhythm, then fine-tuned on binary AF classification using an iterative layer freezing schedule. Deep Taylor decomposition (DTD) was applied to analyze explanations across all configurations. Fine-tuning accuracy ranged from 85.70% to 95.23%, remaining comparable to the original xECGArch architecture and previous TL-based approaches. However, DTD analysis demonstrated that morphology pre-training directed relevance toward P waves, whereas rhythm pre-training concentrated explanations on R peaks, with domain specificity increasing as more layers were frozen. These findings suggest that inductive TL can encourage domain-specific feature attribution in DNNs, improving the alignment of post-hoc explanations with clinically meaningful ECG regions.
26 June 2026
Read appraisal →Physiological measurement
Machine learning-based information flow analysis of ECG signals for long QT syndrome
Objective. Cardiovascular diseases remain the leading cause of death worldwide, highlighting the need for non-invasive and cost-effective risk assessment tools. Biological systems, including the heart, exhibit complex nonlinear dynamics arising from interactions between their subsystems. Information-theoretic measures, particularly entropy-based methods, provide a framework to quantify these interactions. Using ECG recordings, we investigate information flow between heart rhythm and ventricular repolarization to identify potential markers of pathological alterations in cardiac electrical activity.Approach. Entropy-based measures of information transfer were derived from beat-to-beat ECG time series using a window-based approach and subsequently averaged at the subject level. These features were used as inputs to supervised machine learning models to discriminate patients with congenital long QT syndrome (LQTS) from healthy controls. Model performance was evaluated using repeated stratified train-test splits, and classification robustness was assessed across multiple runs using standard performance metrics, including the area under the receiver operating characteristic curve. The explainable artificial intelligence techniques were applied. SHapley Additive exPlanations were used to quantify the contribution of entropy-based features to the model predictions. This post-hoc explainability analysis enabled systematic assessment of feature importance while preserving the predictive performance of the models.Results. The proposed approach achieved high and stable classification performance across repeated validation runs. Both random forest (RF) and support vector machine (SVM) classifiers demonstrated high discrimination between LQTS patients and healthy controls, with consistently high area under the curve. For RF a mean accuracy of 95.9%, mean sensitivity of 95.9%, and mean specificity of 92.9% were achieved across repeated runs. For SVM the corresponding mean values were 93.1%, 93.1%, and 92.0%, respectively.Conclusions.Explainability analysis revealed a dominant contribution of multivariate and conditional information flow features compared with single-source entropy measures, highlighting the relevance of joint and conditional interactions in the classification patterns.
19 May 2026
Read appraisal →IEEE transactions on bio-medical engineering
Extending Multiscale Characterization of Heart Rate Variability via Deep Learning for Mortality Risk Prediction
OBJECTIVE: To improve mortality risk prediction from heart rate variability (HRV) signals by capturing nonlinear scaling patterns often overlooked by traditional linear analyses. METHODS: This study combines detrended moving average (DMA) analysis with convolutional neural networks (CNNs). DMA curves were computed from 2-hour overlapping windows of 24-hour Holter ECG recordings in 916 survivors and 70 nonsurvivors. A CNN was trained to extract features from these curves and benchmarked against models using traditional HRV and clinical features. RESULTS: The CNN achieved an ROC-AUC of 0.72 and an adjusted hazard ratio of 2.129 for daytime recordings, outperforming standard models. Two patient groups emerged based on DMA scaling patterns. Group 1, with dominant short-term scaling, exhibited reduced slopes in nonsurvivors, suggesting impaired autonomic adaptability. Group 2 showed earlier transitions between short- and long-term behavior, where reduced long-term slopes more strongly predicted mortality. Integrated gradients analysis identified key timescales in the DMA curve driving model predictions. CONCLUSION: DMA combined with CNNs enhances HRV-based mortality risk stratification and reveals distinct physiological scaling patterns associated with survival outcomes. SIGNIFICANCE: This study highlights the potential of DMA and CNNs in improving mortality risk stratification and providing mechanistic insights into HRV dynamics, with implications for personalized health monitoring.
4 May 2026
Read appraisal →