Predicting Heart Rate Estimation Accuracy: Photoplethysmography Sensor Measurements on Eight Anatomical Sites

The widespread use of Photoplethysmography (PPG) wearables in various form factors has facilitated long-term continuous cardiac monitoring beyond clinical settings. This study aims to investigate the accuracy of infrared PPG sensors in heart rate (HR) estimation across eight anatomical sites and to characterize the relationship between PPG waveform features and HR estimation accuracy at these sites. We collected PPG data from eight anatomical locations (fingertip, finger base, wrist, forearm, earlobe, mastoid bone, temple, and forehead) to estimate HR. Electrocardiogram (ECG) data were collected and used as ground truth for HR measurement. PPG-based HR error was defined as the normalized difference between PPG- and ECG-based HR values. We quantified the associations between PPG waveform characteristics and HR error using two signal quality metrics: perfusion index (PI) and cardiac power ratio (PR). We found that PI effectively predicted HR errors at anatomical sites with strong pulsatile components (e.g., fingers) but was less reliable at sites with weaker pulsatile characteristics (e.g., wrist). In contrast, PR provided more accurate HR error predictions at sites with weaker pulses. These findings suggest that PR could serve as a more reliable metric for predicting HR errors. Specifically, PR would allow the use of all recordings, enabling users to assess the reliability of reported HR values based on predicted HR error, rather than discarding "low-quality" recordings, as is commonly done with metrics such as PI.Clinical Relevance-In this study, we characterized the relationship between two PPG signal quality metrics (PI and PR) and PPG-based HR estimation errors. These metrics offer insights into the accuracy of HR measurements, enabling clinicians to better assess the reliability and trustworthiness of physiological characteristics across different anatomical sites.