Physiological measurement最新文献

{"title":"A Physiology- and Measurement-Informed Framework for Evaluating PPG Features in Cuffless Blood Pressure Estimation.","authors":"Myung-Kyu Yi, Jongshill Lee, Jeyeon Lee, In Young Kim","doi":"10.1088/1361-6579/aeb063","DOIUrl":"https://doi.org/10.1088/1361-6579/aeb063","url":null,"abstract":"<p><p>Continuous and cuffless blood pressure (BP) estimation using&#xD;photoplethysmography (PPG) has considerable potential for wearable&#xD;health monitoring. However, PPG records peripheral optical&#xD;blood-volume-related changes rather than arterial pressure directly,&#xD;and its waveform is jointly shaped by systemic hemodynamics, local&#xD;vascular regulation, sensor--tissue interaction, and measurement&#xD;conditions. Consequently, high predictive performance does not&#xD;necessarily establish that a model has learned physiologically valid or&#xD;mechanism-specific BP information. To address this gap, we propose a&#xD;physiology- and measurement-informed framework for evaluating&#xD;PPG-derived descriptors and representations. Rather than treating&#xD;features as direct measurements of isolated cardiovascular&#xD;mechanisms, the framework organizes them into four non-exclusive physiological sensitivity domains-cardiac pump-related dynamics, arterial compliance-related dynamics, wave-reflection-related morphology, and peripheral vascular regulation-along with a fifth distinct category for integrative composite representations. Drawing on a qualitative synthesis of 87 studies guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA), representative descriptors are examined in terms of their directly observed signal properties, candidate physiological sensitivities, mechanism specificity, susceptibility to confounding, calibration dependence, wearable observability, and the evidential basis and limitations of their proposed interpretations. Importantly, the supporting literature provides uneven and often indirect validation of the assumed relationships between peripheral blood-volume-related signals, arterial pressure or flow, and specific hemodynamic mechanisms. Particular attention is given to practical limitations such&#xD;as heart-rate dependence, temperature, contact pressure, measurement&#xD;site, and the loss of morphological landmarks in reflective wearable&#xD;PPG. A brief illustrative audit of a published feature-guided hybrid&#xD;model is retained as an example of how the framework can distinguish&#xD;model use of physiologically motivated descriptors from physiological&#xD;validation. The proposed framework does not establish or validate PPG&#xD;as a direct or mechanism-specific measure of BP; rather, it organizes&#xD;the heterogeneous and incompletely validated physiological and&#xD;measurement assumptions reported in the literature. It&#xD;provides a structured basis for identifying conditions under which&#xD;BP-related information may be observable and for defining stronger&#xD;validation requirements for cuffless BP estimation systems.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-10-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149113491","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Dense temporal sleep apnea profiling from single-channel PPG with downstream AHI estimation: a large-scale retrospective feasibility study.","authors":"Filippos Ioannis Katsaridis, Sotirios Fouzas, Ilias Theodorakopoulos","doi":"10.1088/1361-6579/aeafaa","DOIUrl":"https://doi.org/10.1088/1361-6579/aeafaa","url":null,"abstract":"<p><strong>Objective: </strong>Sleep apnea (SA) diagnosis remains challenging due to the cost and limited accessibility of polysomnography. Most automated screening approaches estimate the Apnea-Hypopnea Index (AHI) from oximetry-derived features and do not provide temporally resolved characterization of respiratory events. This study investigates the feasibility of dense temporal respiratory event profiling and clinically meaningful SA severity estimation from single-channel photoplethysmography (PPG).&#xD;Approach: A two-stage framework was developed using single-channel PPG. A fully convolutional neural network operating on PPG-derived waveforms outputs time-resolved probabilities for normal breathing, central apnea, obstructive apnea, hypopnea, and desaturation-only states. These predictions are aggregated via a machine learning metamodel to estimate clinically relevant indices including AHI, Obstructive - Central AHI difference, and the Apnea-to-Apnea-Hypopnea ratio. Models were trained and evaluated on 2029 polysomnography studies from the Multi-Ethnic Study of Atherosclerosis.&#xD;Main results: Fine-grained discrimination between respiratory event subtypes remained challenging, particularly for rare classes like central apnea, whereas aggregated respiratory disturbance detection was more reliable (apnea AUPRC = 0.28, prevalence = 1.9%, AUROC = 0.92). In the PPG-only setting, AHI estimation explained 76% of variance; inclusion of an auxiliary SpO₂-derived desaturation feature improved this to 83%. SpO₂ assisted SA severity classification achieved Macro F1 of 0.71, with only 1.18% of apnea cases misclassified as non-apneic.&#xD;Significance: These findings support the feasibility of dense temporal respiratory disturbance profiling and SA severity estimation from single-channel PPG, while auxiliary oximetry can improve severity estimation when available. They also indicate that fine-grained subtype discrimination remains challenging from PPG alone.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149113433","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"PhysioMerge: a tool for physiological data extraction and analysis.","authors":"Taha Abbasi-Hashemi, Baraa K Al-Khazraji","doi":"10.1088/1361-6579/aeafa8","DOIUrl":"https://doi.org/10.1088/1361-6579/aeafa8","url":null,"abstract":"<p><p>Physiological research relies on accurate data extraction from recording systems such as LabChart. While these systems provide essential tools for data collection, processing large and complex datasets can remain time consuming and often requires manual validation of breaths or cardiac cycles. Manual validation introduces variability due to human judgment, fatigue, and the inability to reprocess large datasets efficiently. To streamline this process, we present PhysioMerge, an automated tool for robust physiological data extraction. PhysioMerge allows users to implement customizable checks to validate signals against predefined criteria, flag anomalies, and preserve valid segments, reducing the need for subjective manual assessment. The software also supports advanced waveform morphology analysis, extracting features such as peak prominence, curvature, and slopes that complement conventional analyses. Flexible data segmentation by time, intervention, or study phase, combined with batch processing capabilities, allows researchers to efficiently handle large datasets and perform consistent, high-precision analyses.&#xD.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149113605","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Class-imbalance-sensitive evaluation of automated decisions in mixed heart-lung auscultation.","authors":"Lulu Wang","doi":"10.1088/1361-6579/aea9eb","DOIUrl":"10.1088/1361-6579/aea9eb","url":null,"abstract":"<p><p><i>Objective.</i>Automated auscultation models are commonly evaluated using aggregate metrics, although mixed heart-lung recordings may be severely imbalanced and contain related acquisition conditions. This study examined whether prevalence, grouping, imbalance correction and decision-threshold selection materially affect the interpretation of cardiac and pulmonary abnormality decisions.<i>Approach.</i>A deterministic log-mel pipeline and nested grouped evaluation based on metadata-defined heart-lung sound-state combinations were applied to 145 mixed HLS-CMDS manikin recordings forming 60 operational groups. Independent and shared convolutional neural networks (CNNs) and a logistic-summary comparator were assessed using five repeated six-fold outer evaluations, with thresholds selected exclusively from inner grouped out-of-fold predictions. All-positive baselines, imbalance-training ablations, preprocessing sensitivity and support requirements for context analysis were evaluated.<i>Results.</i>The deterministic all-positive baselines achieved<i>F</i>1-scores of 0.953 for cardiac detection and 0.893 for pulmonary detection while failing to identify any normal recordings, demonstrating that high<i>F</i>1-score could be prevalence-driven. Across the five repeat-pooled outer evaluations, the logistic-summary comparator achieved the highest pulmonary<i>F</i>1-score, balanced accuracy and Matthews correlation coefficient (MCC), with mean ± SD values of 0.904 ± 0.015, 0.703 ± 0.041 and 0.451 ± 0.077, respectively. The independent CNN achieved the highest pulmonary specificity of 0.643 ± 0.143. Cardiac normal-class recovery remained unstable and did not support robust discrimination. For the independent pulmonary CNN, combined class weighting and oversampling did not improve balanced accuracy or MCC relative to class weighting alone. Support-aware analysis identified insufficient normal-context support for a reliable between-context comparison.<i>Significance.</i>Grouped partitioning, trivial baselines, nested threshold selection and normal-class-sensitive metrics materially changed performance interpretation. The results support methodological safeguards for automated auscultation evaluation but do not establish clinical diagnostic validity.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149002396","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SleepMambaNet: a lightweight multimodal framework for sleep staging and sleep disorder screening with multicenter validation.","authors":"Chao Zhang, Queliang Wang, Yaping Liu, Xiaoliang Li, Fan Jiang, Shuyan Li, Weirong Cui, Jingjing Guo","doi":"10.1088/1361-6579/aeab40","DOIUrl":"https://doi.org/10.1088/1361-6579/aeab40","url":null,"abstract":"<p><p><i>Objective.</i>Manual sleep staging is labor-intensive, whereas many automated systems remain computationally demanding and are validated on a single dataset. This study aimed to develop a lightweight multimodal framework for automated sleep staging and exploratory sleep-disorder screening from polysomnographic (PSG) signals.<i>Approach.</i>SleepMambaNet combines efficient channel attention to integrate complementary information across PSG channels with bidirectional Mamba modules to model long-range temporal dependencies between sleep epochs. The framework was evaluated on the public ISRUC-S1 and ISRUC-S3 datasets and a multicenter clinical cohort of 185 PSG recordings.<i>Main results.</i>In subject-independent experiments, SleepMambaNet achieved accuracy, macro-<i>F</i>1, and Cohen's kappa of 0.852, 0.824, and 0.803 on ISRUC-S3; 0.830, 0.801, and 0.773 on ISRUC-S1; and 0.854, 0.819, and 0.806 on the combined public and clinical data, respectively. The external cross-dataset evaluation yielded accuracy, macro-<i>F</i>1, and kappa of 0.828, 0.812, and 0.779. For exploratory sleep-disorder screening, the model achieved accuracy of 0.892,<i>F</i>1 score of 0.887, sensitivity of 0.956, specificity of 0.842, and precision-recall area under the curve of 0.831.<i>Significance.</i>These results show that cross-channel physiological information and inter-epoch temporal context can be modeled effectively with only 0.47 million trainable parameters across heterogeneous public and clinical PSG data. The combination of computational efficiency and cross-dataset performance provides a practical basis for prospective evaluation of automated PSG decision support, while the binary screening output should not be interpreted as a disease-specific or standalone clinical diagnosis.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":"47 10","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149113525","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Single-lead ECG biomarkers for cardiovascular and mortality risk prediction.","authors":"Lisa Attali, Yosef Solewicz, Shany Brimer Biton, Hagai Hamami, Eran Zvuloni, Ilan Green, Izhar Laufer, Pablo Laguna, Alba Martín-Yebra, Juan Pablo Martínez, Ronit Almog, Joachim A Behar","doi":"10.1088/1361-6579/aea9ec","DOIUrl":"10.1088/1361-6579/aea9ec","url":null,"abstract":"<p><p><i>Objective.</i>Cardiovascular disease remains one of the leading causes of morbidity and mortality worldwide, highlighting the need for accurate risk prediction. Atrial fibrillation burden (AFB), premature ventricular contraction burden (PVCB), and T-wave alternans (TWA) have been individually associated with adverse outcomes. We hypothesized that these three digital ECG biomarkers provide complementary prognostic information for major cardiovascular endpoints and all-cause mortality (ACM), and developed a machine learning approach combining these biomarkers for risk prediction.<i>Approach.</i>We analyzed 81 362 Holter recordings from 54 395 individuals across 20 primary care centers in Israel. A random forest model using AFB, PVCB, TWA, and age was trained to predict 5 year risk of heart failure (HF), ischemic stroke (IS), and ACM.<i>Main results.</i>On the test set, best AUROCs were for HF 0.75 [95% CI: 0.74-0.77] (), for IS 0.69 [0.66-0.71] (), and for ACM 0.79 [0.77-0.80] (). Combining the three ECG-derived biomarkers showed complementary predictive value and improved discrimination by up to 10% over age alone in individuals agedyears.<i>Significance.</i>AFB, PVCB, and TWA contribute complementary prognostic information for risk stratification. When combined with age, these biomarkers improve predictive performance compared with age alone.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149002313","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Prediction of cerebral palsy in extremely preterm infants with beat-to-beat and lower-resolution heart rate variability metrics.","authors":"R B Govindan, Venkata Chaitanya Chirumamilla, Lisa Letzkus, Adre du Plessis, Karen D Fairchild, Douglas E Lake","doi":"10.1088/1361-6579/aea9e9","DOIUrl":"10.1088/1361-6579/aea9e9","url":null,"abstract":"<p><p><i>Objective.</i>To validate previous findings that low heart rate variability (HRV) in the first week after birth is associated increased risk of cerebral palsy (CP). Additionally, to quantify the extent to which higher resolution beat-to-beat HRV metrics improve CP prediction and determine the effect of heartbeat averaging as done in bedside monitors.<i>Approach.</i>For infants <29 weeks gestation at a level IV NICU, beat-to-beat heart rate (btbHR) and corresponding RR intervals (btbRR) from standard monitors were retrieved from a data repository. Monitor-displayed HR based on 12-beat averaging and measured at 1 Hz was also collected. The data were segmented into 10 min windows, and multiple HRV metrics were calculated including standard deviation (SD), low- and high-frequency HRV (LF, HF) and normalized low- and high-frequency HRV. Additionally, detrended fluctuation analysis (DFA) slope (alpha) and root mean square (RMS) metrics were calculated for both short (alpha1, RMS1) and long range (alpha2, RMS2) fractal scaling exponents. Metrics were averaged over the first week after birth for each infant to provide representative values that were used to quantify CP prediction performance using the area under the receiver operating characteristic curve (AUC).<i>Main Results.</i>This study included 102 infants,11 (10.8%) with CP. The SD of the 1 Hz monitor HR had AUC 0.754 compared to the highest values of 0.769 and 0.764 for DFA RMS1 and RMS2. Since raw btbHR contains erroneous beats, averaging generally improves performance. With averaging over 12 beats, as is done by most bedside monitors, RMS2 of btbHR and LF of btbRR had highest AUCs of 0.764 and 0.771 respectively.<i>Significance.</i>We externally validated that, in preterm infants during the first week after birth, reduced HRV predicts subsequent diagnosis of CP. Measures calculated from btbHR data showed slightly superior performance compared with 1 Hz heart rate data displayed on standard NICU bedside monitors.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149002399","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A gated artifact management pipeline for low-density eyewear EEG.","authors":"Andrea Costanzo Palmisciano, Andrea Farabbi, Francesco Latino, Matteo Rossi, Niccoló Antonello, Pietro Cerveri, Luca Mainardi","doi":"10.1088/1361-6579/ae99ab","DOIUrl":"10.1088/1361-6579/ae99ab","url":null,"abstract":"<p><p><i>Objective</i>. To develop an artifact-management method suitable for low-density, dry-electrode electroencephalogram (EEG) recorded from a wearable eyewear platform, enabling artifact detection, classification, and correction while avoiding unnecessary processing of clean neural activity.<i>Approach</i>. A gated two-stage pipeline is proposed in which a lightweight gate classifier first identifies artifact-contaminated EEG windows and selectively triggers further processing only when needed. The complete pipeline combines: (i) a time-domain feature-based detector, (ii) a convolutional-recurrent network for three-class artifact discrimination (eye blinks, horizontal eye movements, facial movements), and (iii) a U-Net-based denoising autoencoder.<i>Main results</i>. Evaluated on the eyewear EEG dataset (29 subjects) using a leave-one-subject-out scheme, the gate classifier achieved a balanced accuracy of 0.89. The artifact classifier reached a balanced accuracy of 0.84 with a median macro-F1 of 0.83. The denoiser achieved a median normalized RMSE lower than 0.20 and a spectral cosine similarity above 0.84. In the end-to-end evaluation, the pipeline achieved a median normalized RMSE of 0.19 and a spectral cosine similarity of 0.89. Evaluation on an independent external benchmark showed that the proposed pipeline maintained a slightly lower gate detection performance (balanced accuracy 0.82) and achieved higher two-class classification performance. The denoising performance was comparable to that obtained on the dataset collected for this study, with absolute median differences not exceeding 0.04 for both normalized RMSE and spectral cosine similarity.<i>Significance</i>. These results support the effectiveness of the proposed approach for artifact management in wearable eyewear EEG, while maintaining a design compatible with resource-constrained wearable implementations.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148764532","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Measurement properties and operational feasibility of an automated arterial occlusion pressure assessment system for blood flow restriction exercise.","authors":"Joanderson Felipe Soares Silva, Nicholas Rolnick, Maria Luiza Araujo Cerqueira, Gabriela Mota Pinto Silva, Rafael Pereira, Mikhail Santos Cerqueira","doi":"10.1088/1361-6579/aea9ed","DOIUrl":"10.1088/1361-6579/aea9ed","url":null,"abstract":"<p><p><i>Objective.</i>To evaluate the validity, reliability, agreement, and operational feasibility of an automated arterial occlusion pressure (AOP) assessment system compared with a manual Doppler-based assessment.<i>Approach.</i>Forty healthy adults (20 men and 20 women) completed upper- and lower-limb (LL) AOP assessments using automated and MDA methods in a randomized and counterbalanced design. Concurrent validity, agreement, intraday and interday reliability, operational feasibility, perceptual responses, and procedure-related symptoms were evaluated. Exploratory analyses examined whether sex modified the difference between automated and MDA-derived AOP measurements.<i>Main results.</i>MDA demonstrated good-to-excellent reliability (ICC = 0.84-0.98; CV = 1.38%-3.25%). The automated system demonstrated moderate-to-good intraday reliability (ICC = 0.74-0.85) but poor LL interday reliability (ICC = 0.36). Compared with MDA, automated AOP showed systematic bias and wide limits of agreement. No evidence of a sex-related difference in the between-method discrepancy was observed for either upper- or LL AOP. Although automated assessment reduced testing time by 81.9%-86.7%, frequent upper-limb (UL) measurement failures reduced the automated measurement success rate to approximately 57%, substantially increasing adjusted assessment time and limiting operational feasibility, while discomfort and affective responses were more favorable than with MDA.<i>Significance.</i>The automated AOP system improved assessment efficiency and participant experience but demonstrated less favorable reliability estimates, limited agreement, and frequent upper-limb measurement failures. Consequently, automated and MDA should not be considered interchangeable for individualized AOP prescription.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149002388","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A tri-modal, offline, and explainable framework for 12-lead electrocardiogram interpretation: fusing deterministic rules, a local residual network, and an agentic vision-language model.","authors":"Mohammad Mahbubur Rahman Khan Mamun","doi":"10.1088/1361-6579/aeaded","DOIUrl":"https://doi.org/10.1088/1361-6579/aeaded","url":null,"abstract":"<p><strong>Objective: </strong>Automated electrocardiogram (ECG) interpretation is increasingly delivered as cloud deep-learning services, trading data sovereignty, auditability and robustness for accuracy. This study tests whether an offline, explainable framework can remove that trade-off.&#xD;&#xD;Approach. Three independent interpreters feed a decision-fusion layer: a deterministic engine of 26 executable clinical criteria, each finding linked to a highlighted signal segment and a literature source; a compact one-dimensional residual network trained on PTB-XL and run on-device via the Open Neural Network Exchange runtime; and an optional agentic vision-language model reading a rendered ECG image behind a security screen with human-in-the-loop escalation. Format-agnostic ingestion, lead-aware capability tiering and a three-tier signal-quality gate precede interpretation. Evaluation covers diagnostic accuracy, on-device latency, noise resilience, modality ablation and zero-shot external validation.&#xD;&#xD;Main results. On the PTB-XL test fold (n = 2198) the network reaches a macro-averaged area under the receiver operating characteristic curve (AUROC) of 0.932 (95% confidence interval (CI) 0.918-0.943), matching published GPU baselines, while the complete pipeline runs in 108 ± 25 ms on a CPU with no network access. Performance degrades gracefully to a signal-to-noise ratio of 5 dB (AUROC 0.951) as the quality gate rejects uninterpretable signals, and rules and network agree only weakly (Cohen's kappa 0.26), indicating complementary errors. Applied without retraining to an independent cohort (Chapman-Shaoxing, n = 1599) the model retains a macro-AUROC of 0.898. In a small pilot (25 records; 30 rule/network disagreements) the general-purpose agentic model was the weakest adjudicator (accuracy 0.267, 95% CI 0.142-0.444; exact McNemar test versus rules, p = 0.09).&#xD;&#xD;Significance. Edge inference, intrinsic explainability and governed multi-modal fusion give competitive accuracy with auditable findings. The agentic layer is positioned for governance rather than diagnosis, motivating ECG-specialized models. The open-source system is a research prototype, not a medical device.</p>","PeriodicalId":20047,"journal":{"name":"Physiological measurement","volume":" ","pages":""},"PeriodicalIF":2.5,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149081812","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}