A photoplethysmogram-based 1D-CNN algorithm for automated atrial fibrillation detection

Atrial fibrillation (AF) is an irregular heartbeat that originates in the upper chambers of the heart (atria). This paper presents a method for the detection of AF events from photoplethysmogram (PPG) signals, which are the sensing standard for heart rate and oxygen saturation monitoring in wearable devices such as smartwatches, contrasting traditional diagnostic methods based on electro-cardiograms. The paper details an AF detection algorithm using a 1-dimensional convolutional neural network developed utilizing the MIMIC III Waveform database. The developed model achieved an accuracy of 98.27%, and an F1-score of 97.78%, outperforming several state-of-the-art PPG-based AF detection methods. Additionally, the feasibility of model pruning and binarization to reduce computational complexity is explored. The pruned network (60% sparsity) achieved 97.26% accuracy and a 96.40% F1-score, while the binarized network attained 94.51% accuracy and a 93.29% F1-score. The performance of the proposed method is compared against state-of-the-art algorithms.Clinical relevanceThis study introduces a highly accurate deep learning-based method for AF detection using PPG signals, enabling reliable diagnosis through wearable devices. By facilitating continuous telemonitoring, this approach enhances early AF detection, reduces the burden on healthcare practitioners, and improves patient outcomes.