Improving Photoplethysmography Peak Detection through Higher-Order Interpolation Techniques

Maissa Maniezzo1, Arthur Valencio1, Italo Sandoval Ramos de Oliveira2, Rafael Lima1
1Samsung R&D Institute Brazil, 2Samsung R&D Brazil


Abstract

Photoplethysmography (PPG) is an optical technique widely used in wearable devices for non-invasive monitoring of cardiovascular health. Nonetheless, accurate detection of PPG peaks remains a significant challenge due to noise and signal variability that impacts the reliability and effectiveness of wearable health technologies. Inaccurate measurements can lead to misdiagnosis or delayed diagnosis of cardiovascular conditions. This paper presents a novel approach to improve PPG peak detection through the addition of a polynomial interpolation module with degree greater than 2, aiming to best reproduce the PPG true waveform compared to Fiovaranti et al. Polynomial solutions of orders 3 to 5 have been considered and compared to order 2 (Fioravanti et al's method) and without peak correction. The optimization method of the polynomial is also considered, by the determination of its coefficients through different standard optimizers. Our evaluation of higher-order polynomial interpolators demonstrate improved accuracy, precision and recall compared to rule-based peak detection algorithms from literature such as HeartPy and Elgendi. Comparing the polynomial degrees, we found the best metrics coming from order 3, expressing a gain from Fiovaranti et al original proposal. The marginal increase in accuracy can have profound effects in Interbeat Intervals applications, such as arrhythmia detection, as it represents a decrease in false cases of missed beats from PPG peaks compared to ECG peaks, which is the clinical reference.