Key result
Novel machine learning removes artefacts from PPG-derived HR data, achieving high agreement with reference activity scores.
Why the study?
PPG-based heart rate monitoring enables continuous physical activity assessment, but device-generated values are prone to motion artefacts and raw waveforms are unavailable in commercial devices.
Does a novel machine learning procedure applied to PPG-derived heart rate data improve the accuracy of physical activity assessment compared to a reference chest strap?
Cohort (n=149)
Does a novel machine learning procedure applied to PPG-derived heart rate data improve the accuracy of physical activity assessment compared to a reference chest strap?
Effect estimate: Pearson's r 0.89-0.96
A novel machine learning procedure applied directly to device-generated PPG heart rate data effectively removes motion artefacts and yields high agreement with reference chest strap data for physical activity assessment.
Supports reliable PPG-based activity assessment in clinical practice; extends validation of machine learning artefact rejection to real-world cohorts.
Introduction Photoplethysmography (PPG) based heart rate (HR) monitoring supports continuous assessment of physical activity intensity, but device generated HR values remain prone to motion artefacts. This study validated a novel machine learning procedure that detects artefacts directly in PPG derived HR data, rather than in raw waveforms that are unavailable in commercial devices. Methods In this prospective study, 149 participants (46 following cardiac rehabilitation, 103 healthy) wore a PPG-based wrist device and a reference chest strap for 12 weeks. Prior testing defined participants as “PPG-compatible” (i.e., HR error <10% during ≥70% of training data). Three quarters trained artefact and activity detection models, one quarter served as the test population. To balance artefact removal and activity preservation, multiple probability thresholds were combined to reject unreliable HR values, remove or interpolate using adjacent reliable values. The Antwerp Activity Index (AAI), an HR-based PA score, was calculated from the resulting PPG and reference data to evaluate threshold combinations. Results Seventy-eight participants (53.8%) were PPG-compatible, yielding over 5 million datapoints, with 75% ( n = 58; 4,144,654 datapoints) used for training and 25% ( n = 20; 992,180 datapoints) for testing. The artefact model detected artefacts with a sensitivity of 58.5% in daily life and 76.6% during exercise, while limiting incorrect removals (74.9–91.3% specificity). Rejecting data for AAI-calculation when artefact probability was >50% and activity probability <70% produced the highest agreement with reference AAI (Pearson’s r = . 89–.96). Conclusions This new machine learning-based procedure, which operates on device-generated HR rather than PPG waveforms, effectively removes artefacts while preserving key activity signals, ensuring clinically relevant PA assessment in daily life.
No takes yet. Share an insight, caveat, or question.
Vermunicht et al. (2026) conducted a cohort in Cardiac rehabilitation and healthy (n=149). Machine learning procedure for artefact detection vs. Reference chest strap was evaluated on Agreement with reference Antwerp Activity Index (AAI) (Pearson's r 0.89-0.96). A novel machine learning procedure operating on device-generated HR data effectively removed artefacts, achieving high agreement with reference activity scores (Pearson's r = 0.89-0.96).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: