Causal Peak Refinement Reduces Timing Error of a Classical R-Peak Detector Across Beat Morphologies
Year: 2026
Type: preprint
Abstract: Purpose Classical ECG R-peak detectors are usually reported with a single aggregate score, which can hide failures on unusual beats and under noise. One such detector was evaluated by beat type, noise type and matching tolerance, and a refinement stage was tested. Methods A simplified Pan-Tompkins-style detector and a causal peak-refinement stage, which reassigns each detection to the largest absolute-amplitude sample of the 5–15 Hz band-passed signal within ± 50 ms, were implemented in Simulink. Nine MIT-BIH Arrhythmia Database records (first 120 s each) were used: five to choose settings and four held out. Noise from the MIT-BIH Noise Stress Test Database was added at 24, 12, 6 and 0 dB. Results At ± 50 ms the baseline’s mean F1 was 92.5% (88.1% on a right bundle branch block record and 63.4% on a paced record), but it was 98.2% at ± 75 ms, indicating mainly timing errors. Refinement reduced the mean absolute timing error from 25.8 ms to 6.0 ms and raised the ± 50 ms mean F1 to 98.0%. Refined F1 was equal or higher on all four held-out records. Electrode motion at 0 dB was the hardest noise (tuning records: mean F1 87.3% baseline, 91.7% refined). Refinement did not help record 207, which contains ventricular flutter and unmatched premature ventricular contractions (F1 86.3%). Conclusion Most errors under the ± 50 ms matching criterion were timing errors, and a simple causal refinement stage reduced them. The results come from nine short records and one detector, and they do not rank detectors.
Source: Research Square
Authors: Sadekatul Shihab, Akib Ikbal
Institution: Varendra University
Language: English
Topic: ECG Monitoring and Analysis
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7233545848
