Low-cost ECG devices offer a scalable solution for home-based cardiac monitoring, but their signal fidelity and diagnostic reliability relative to clinical-grade systems remain insufficiently validated. This study evaluates the performance of an in-house low-cost ECG prototype (MCU + amplifier) against a reference system (Norav1200HR).
A total of 528 pooled beats from a single-lead ECG of 22 healthy human participants show excellent morphological similarity to a reference with a median of 0.982 (IQR 0.027) and a near-zero DTW distance of 0.00191. MATLAB R peak detection and QRS onset achieved an MAE of 9.04ms (bias −0.03ms; 95% LoA ± 42.6ms), and 9.59ms, respectively. Both satisfy their respective IEC thresholds of 10ms and 15ms. P-wave onset detection was suboptimal, with an MAE of 28.5ms (bias −6.5 ms; 95% LoA −100.7 to +87.7ms), failing the IEC 15ms threshold, which is consistent with the inherent difficulty of resolving low-amplitude atrial activity in single-lead systems. T-wave end showed moderate performance at an MAE of 20.0ms with a slight positive bias. Heart rate and QRS duration met IEC tolerances, with HR bias of 0.059 bpm (LoA ±3.2 bpm) and QRS bias of −0.030ms (LoA ±12.1 ms). Systematic amplitude biases were identified in S-wave and T-wave measurements (−76.8 μV and +99.7 μV, respectively), likely due to a high-pass filter, frequency mismatch between the prototype and the Norav 1200HR. Beat classification across four classes: Normal, Supraventricular ectopic, Ventricular ectopic, and Noise, achieved 97.0% overall accuracy; however, MATLAB inter-rater reliability of 0.482 indicates moderate agreement after accounting for class imbalance, with reduced sensitivity for ventricular ectopic beats (70.8%).
This prototype demonstrates strong performance in heart rate, QRS delineation, and rhythm analysis, meeting IEC compliance thus, supporting its use in remote monitoring and screening. However, improvements in atrial signal resolution, ventricular ectopic detection, and filter alignment are required for diagnostic-grade applications.