Key result
FeatureDB accurately computes single-lead ECG parameters with ~8.5 ms PR interval error vs experts.
Why the study?
Comprehensive validation of on-device computational methods for continuous ECG parameter measurement across heterogeneous real-world populations remains limited.
Does the FeatureDB algorithm accurately compute ECG parameters and detect cardiac abnormalities from single-lead ECGs compared to expert annotations and 12-lead systems?
Observational (n=21,723)
Yes
Does the FeatureDB algorithm accurately compute ECG parameters and detect cardiac abnormalities from single-lead ECGs compared to expert annotations and 12-lead systems?
The FeatureDB algorithm enables accurate, real-time computation of ECG parameters from single-lead devices, offering a scalable tool for decentralized cardiac screening despite lower diagnostic sensitivity than 12-lead systems.
Supports on-device ECG validation in real-world cohorts; leaves open clinical adoption pending larger prospective trials.
Accurate and continuous electrocardiogram (ECG) parameter measurement outside hospital environments is essential for real-time cardiac health monitoring and telemedicine, yet comprehensive validation of on-device computational methods across heterogeneous real-world populations remains limited. We conducted a real-world validation using two datasets: HeartVoice-ECG-lite (369 participants with single-lead ECG recordings annotated by two cardiologists) and PTB-XL/PTB-XL+ (21,354 patients with 12-lead ECG recordings and diagnostic annotations). FeatureDB ( https://github.com/PKUDigitalHealth/FeatureDB ) was applied to compute PR, QT, and QTc intervals from single-lead signals. Accuracy was assessed using mean absolute error (MAE), correlation, and Bland-Altman analysis. Diagnostic performance for first-degree atrioventricular block (AVBI, based on PR) and long QT syndrome (LQT, based on QTc) was benchmarked against commercial 12-lead systems (12SL, Uni-G) and an open-source algorithm (Deli) using AUC, accuracy, sensitivity, and specificity. FeatureDB-derived parameters showed high concordance with annotations, with MAEs comparable to inter-observer variability and Pearson correlations ranging from 0.836 to 0.960. For AVBI detection, FeatureDB achieved an AUC of 0.787; for LQT, an AUC of 0.684. These results indicate that FeatureDB enables accurate real-time on-device ECG parameter computation suitable for scalable telemedicine, decentralized screening, and continuous community monitoring.
No takes yet. Share an insight, caveat, or question.
Fan et al. (2026) conducted an observational in Cardiac abnormalities (n=21,723). FeatureDB algorithm vs. Expert cardiologist annotations and 12-lead clinical systems was evaluated on Mean absolute error (MAE) of PR interval compared to Doctor A (MAE 8.46 ms, 95% CI 7.64-9.36). The FeatureDB algorithm accurately computed single-lead ECG parameters with a PR interval mean absolute error of 8.46 ms compared to experts, and achieved an AUC of 0.787 for detecting first-degree atrioventricular block.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: