Photoplethysmography (PPG) is a rising technology that provides a simple, affordable, and non-invasive way to continuously monitor the vascular system. Its popularity is largely due to its integration into user-friendly smart devices, which are accessible and affordable. Given the rapid expansion of smart devices, there is a significant opportunity to extend health monitoring to a broader population.
This paper addresses the critical issue of cardiac arrhythmia (CA) detection using PPG signals. CA conditions present significant health risks, often resulting in complications such as stroke and heart failure. While many studies concentrate solely on identifying atrial fibrillation (AF), our research aims to categorize six different rhythm types into three classes: Sinus rhythm, AF, and Other (including premature atrial contraction, premature ventricular contraction, ventricular tachycardia, and supraventricular tachycardia).
For this purpose, 46,827 10s PPG signals from 91 individuals were used. A total of 49 features were extracted from these signals, encompassing not only pulse interval features, but also statistical and morphological features. The entire dataset was divided into training, validation, and test sets, followed by standardization. Twelve features were selected using the forward feature selection method. The Random Forest model with optimal hyperparameters was trained, achieving an F1 score of 0.88 on the test set. Additionally, the model was tested on an independent dataset, achieving an F1 score of 0.87. This study demonstrates the efficacy of the proposed approach in detecting various types of CAs beyond AF, providing valuable insights for enhanced cardiovascular health monitoring.