AI-Based Atrial Fibrillation Detection Using Pulse Rate Variability from PPG: A Comparative Study with HRV

Atrial fibrillation is one of the most prevalent cardiac arrhythmias for an early diagnosis of its effective treatment. Traditionally, the diagnosis of AFib relies on Heart Rate Variability from electrocardiography. However, with the increase in accessibility of Photoplethysmography, Pulse Rate Variability derived from PPG may turn out to be the promising alternative to HRV. This paper investigates the diagnostic efficacy of PRV for AFib detection, comparing it to HRV using statistical analysis as well as several machine learning models. The dataset used is MIMIC-III PERform AF which comprises ECG and PPG recordings. Parameters related to HRV and PRV were calculated by extracting RR intervals from ECG signals and PP intervals from PPG signals. The similarity of HRV and PRV was analyzed through T-test and correlation. Machine learning models like Support Vector Machine SVM and Logistic Regression were trained on features of PRV to classify AFib and non-AFib cases. Results showed a high correlation between the two signals. High accuracy in the classification of AFib using PRV was obtained with machine learning models that were trained on PRV features, and Logistic Regression attained an accuracy of 98%. Results prove that PRV may be an alternative to HRV for AFib detection, especially in wearable applications, as a non-invasive low-cost solution.