Managing of artifacts in cardiovascular time series from photoplethysmographic signals

Photoplethysmography (PPG) is widely used in clinical and consumer devices for its non-invasive and cost-effective nature. PPG signals from wearable devices can provide valuable insights into cardiovascular and respiratory functions in the real-world. However, their accuracy depends on signal quality, which decreases while moving. Several methods, including rule-based approaches and machine learning techniques, have been proposed for detecting noise-corrupted segments.This study systematically reviewed (n = 51) and benchmarked (n = 5) algorithms for artifact detection in PPG signals. A benchmark dataset was created from the publicly available PPG-DaLiA dataset, and algorithms were evaluated using a leave-one-subject-out cross-validation. A novel algorithm was implemented, based on the correlation between PPG beats and a template derived from "artifact-free" beats. Key performance metrics, such as accuracy, sensitivity, specificity, precision, F1-score, and Cohen’s kappa were used to compare algorithms. The systematic review revealed that 57% of the articles used proprietary datasets, over 76% of algorithms relied solely on PPG signals as input, and 73% detected artifact in time-windows of fixed length. Most algorithms (86%) produced binary outputs, indicating the presence or absence of artifacts. Only 16% of algorithms were open source, which were included in the benchmark study. The benchmark showed that machine learning-based algorithms achieve the highest performance: Tiny-PPG achieved the highest specificity (0.8701 ± 0.0448), while the pulse-SVM the highest sensitivity (0.9521 ± 0.0154). Segade demonstrated the best balance between sensitivity and precision, with an F1 score of (0.8864 ± 0.0393). This study lays the foundation for the reliable extraction of heart rate and heart rate variability parameters from real-world PPG data, enabling continuous and unobtrusive monitoring of cardiovascular health using wearable devices.