Convolutional Neural Network-based ECG Classification on PYNQ-Z2 Framework

Abstract

An electrocardiogram (ECG) contains vital information to diagnose and monitor conditions affecting the heart. There are many types of ECG signals depicting heart conditions. Manual classification of these vital signals is an error-prone and time-consuming process. Furthermore, accurate and automated online classification of ECG signals on an Edge platform has been a challenging task owing to the complexity of the inference models. In this paper, we propose a deep convolutional neural network-based method to classify five classes of ECG beats. The model is deployed on a PYNQ-Z2 board. The Analog Discovery Kit reproduces a bioelectrical representation of ECG for each class. The classification task is performed using the proposed model on the PYNQ-Z2 board to achieve an accuracy of 95.6% and an F1-score of 95.61% with lesser parameters. The proposed architecture shows improved performance as compared to state-of-the-art models.