On-device Computation of Single-lead ECG Parameters for Real-time Remote Cardiac Health Assessment: A Real-world Validation Study

Sumei Fan, Deyun Zhang, Yue Wang, Shijia Geng, Kun Lu, M Sang, Weixiang Xu, Hua Wang, Qinghao Zhao, Chuandong Cheng, Wang Peng, Shenda Hong

Abstract

Accurate, continuous out-of-hospital electrocardiogram (ECG) parameter measurement is vital for real-time cardiac health monitoring and telemedicine. On-device computation of single-lead ECG parameters enables timely assessment without reliance on centralized data processing, advancing personalized, ubiquitous cardiac care—yet comprehensive validation across heterogeneous real-world populations remains limited. This study validated the on-device algorithm FeatureDB using two datasets: HeartVoice-ECG-lite (369 participants with single-lead ECGs annotated by two physicians) and PTB-XL/PTB-XL+ (21,354 patients with 12-lead ECGs and physicians' diagnostic annotations). FeatureDB computed PR, QT, and QTc intervals, with accuracy evaluated against physician annotations via mean absolute error (MAE), correlation analysis, and Bland-Altman analysis. Diagnostic performance for first-degree atrioventricular block (AVBI, PR-based) and long QT syndrome (LQT, QTc-based) was benchmarked against commercial 12-lead systems (12SL, Uni-G) and open-source algorithm Deli, using AUC, accuracy, sensitivity, and specificity. Results showed high concordance with expert annotations (Pearson correlations: 0.836-0.960), MAEs matching inter-observer variability, and minimal bias. AVBI AUC reached 0.787 (12SL: 0.859; Uni-G: 0.812; Deli: 0.501); LQT AUC was 0.684 (12SL: 0.716; Uni-G: 0.605; Deli: 0.569)—comparable to commercial tools and superior to open-source alternatives. FeatureDB delivers physician-level parameter accuracy and commercial-grade abnormality detection via single-lead devices, supporting scalable telemedicine, decentralized cardiac screening, and continuous monitoring in community and outpatient settings.

Introduction

While single-lead wearable ECGs have democratized basic cardiac monitoring, their clinical utility is often limited to simple heart rate tracking. Diagnosing complex conditions like conduction blocks or drug-induced QT prolongation requires precise measurement of specific waveform intervals (PR, QRS, QT). Extracting these parameters from noisy wearable signals is algorithmically challenging, and cloud-based processing introduces latency and privacy risks. This study addresses these challenges by validating FeatureDB, an on-device (edge computing) algorithm designed for clinical-grade ECG parameter extraction from single-lead devices.

Methods and Validation Framework

FeatureDB utilizes a multi-stage pipeline involving robust peak detection and multi-scale wavelet transforms to delineate wave boundaries (P wave, QRS complex, T wave). The algorithm was rigorously tested using a dual-dataset approach:

  1. HeartVoice-ECG-lite: A real-world dataset of 369 participants using single-lead wearables, manually annotated by two independent cardiologists to establish a human baseline.
  2. PTB-XL/PTB-XL+: A massive clinical benchmark of 21,354 patients with 12-lead ECGs, used to compare FeatureDB against industry-standard commercial algorithms (GE 12SL, Glasgow Uni-G) and open-source alternatives (Deli).
Figure 1: Overview of FeatureDB development, dataset construction, and the comparative analysis framework.

Key Results

Parameter Accuracy vs. Human Experts

FeatureDB demonstrated exceptional concordance with physician annotations. Pearson correlation coefficients ranged from 0.836 to 0.960 for PR, QT, and QTc intervals. Bland-Altman analysis confirmed minimal bias, with error rates comparable to the natural inter-observer variability between two human cardiologists. This suggests that edge algorithms can reliably perform complex waveform delineation previously reserved for human experts.

Diagnostic Performance

The algorithm's ability to detect cardiac abnormalities was benchmarked against gold-standard 12-lead systems:

  • AV Block (AVBI): FeatureDB achieved an AUC of 0.787, significantly outperforming the open-source Deli algorithm (0.501) and remaining competitive with commercial 12-lead systems (12SL: 0.859; Uni-G: 0.812).
  • Long QT Syndrome (LQT): FeatureDB achieved an AUC of 0.684, outperforming the commercial Uni-G system (0.605) and trailing only the industry-standard 12SL (0.716). Notably, FeatureDB exhibited the highest sensitivity (0.617) among all tested methods.
Figure 2: Performance metrics (AUC, Accuracy, Sensitivity, Specificity) for AVBI and LQT detection across different algorithms.

Discussion and Future Outlook

This validation proves that on-device, single-lead ECG analysis can achieve physician-level accuracy and commercial-grade diagnostic performance. By moving computation to the edge, FeatureDB enables real-time, privacy-preserving monitoring in outpatient settings, which is critical for patients on high-risk medications or in underserved regions. Future work will focus on expanding demographic representation and evaluating the long-term impact of decentralized monitoring on clinical outcomes.

Summary updated

Abstract

Accurate, continuous out-of-hospital electrocardiogram (ECG) parameter measurement is vital for real-time cardiac health monitoring and telemedicine. On-device computation of single-lead ECG parameters enables timely assessment without reliance on centralized data processing, advancing personalized, ubiquitous cardiac care-yet comprehensive validation across heterogeneous real-world populations remains limited. This study validated the on-device algorithm FeatureDB (https://github.com/PKUDigitalHealth/FeatureDB) using two datasets: HeartVoice-ECG-lite (369 participants with single-lead ECGs annotated by two physicians) and PTB-XL/PTB-XL+ (21,354 patients with 12-lead ECGs and physicians' diagnostic annotations). FeatureDB computed PR, QT, and QTc intervals, with accuracy evaluated against physician annotations via mean absolute error (MAE), correlation analysis, and Bland-Altman analysis. Diagnostic performance for first-degree atrioventricular block (AVBI, PR-based) and long QT syndrome (LQT, QTc-based) was benchmarked against commercial 12-lead systems (12SL, Uni-G) and open-source algorithm Deli, using AUC, accuracy, sensitivity, and specificity. Results showed high concordance with expert annotations (Pearson correlations: 0.836-0.960), MAEs matching inter-observer variability, and minimal bias. AVBI AUC reached 0.787 (12SL: 0.859; Uni-G: 0.812; Deli: 0.501); LQT AUC was 0.684 (12SL: 0.716; Uni-G: 0.605; Deli: 0.569)-comparable to commercial tools and superior to open-source alternatives. FeatureDB delivers physician-level parameter accuracy and commercial-grade abnormality detection via single-lead devices, supporting scalable telemedicine, decentralized cardiac screening, and continuous monitoring in community and outpatient settings.
BibTeX
@misc{fan2025device,
  author = {Sumei Fan and Deyun Zhang and Yue Wang and Shijia Geng and Kun Lu and M Sang and Weixiang Xu and Hua Wang and Qinghao Zhao and Chuandong Cheng and Wang Peng and Shenda Hong},
  title = {{On-device Computation of Single-lead ECG Parameters for Real-time Remote Cardiac Health Assessment: A Real-world Validation Study}},
  year = {2025},
  eprint = {2502.17499},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2502.17499},
}
Figure