Wireless Electrocardiograph Monitoring Based on Wavelet Convolutional Neural Network

Recently, cardiovascular diseases (CVD) have drawn high concerns from diversity of disciplines due to acute and fatal characteristics. Rapidly accumulating evidence suggests that CVD threats human health without any symptom. Finding a way to forestall it or notice ongoing warning becomes significant. Electrocardiograph (ECG) monitoring is one of the commonly used techniques to address this problem. The aim of this project is to explore various deep learning models with data collected from our developed wearable ECG patch: IREALCARE. We applied new dataset and tested the performance of existing Convolutional Neural Network (CNN) and Long Short-term Memory (LSTM) models. However, current models are not able to identify two main classes (class S and A), resulting in 58.0% mean accuracy. To overcome this drawback, a Wavelet Based Convolutional Neural Network (WBCNN) is proposed and the final mean accuracy reaches 77.7%.