Cardiac Arrhythmia Detection Leveraging Deep Learning Convolutional Neural Network

Cardiac arrhythmia is increasingly prevalent and affected patients have an increased lifetime risk of stroke and heart failure. The aim of this study is to develop a deep convolutional neural network (DCNN) capable of differentiating atrial fibrillation (AF) from sinus rhythm (SR) and premature contractions and using a photoplethysmography (PPG) signal. 46827 10-s segments of raw PPG-signal data from 91 patients is used to train a 1D VGG16-Net model to predict the rhythm of the segments. Overall performance from 10-fold cross validation reach an balanced accuracy of 0.895 ± 0.03, and testing on independent data reach a balanced accuracy of 0.8279. The DCNN proposed in this study is capable of differentiating between SR, premature contractions and AF, but there is a need to advance model accuracy.Clinical Relevance - This further legitimizes the potential use of PPG signal data in screening for atrial fibrillation, but also for premature contractions.