Intelligence ECG Monitoring System: Wireless platform and Arrhythmia classification using Residual neural network

Abstract

The real-time wireless 6-lead electrocardiogram (ECG) monitoring platform with the application of a convolutional neural network (CNN) in Arrhythmia classification is proposed in this study. The platform consists of two main parts. The mainboard with ADS1292R: a 24-bit ECG analog frontend integrated circuit and a low-cost low-power microcontroller: ESP32 which received and resample to 250 Hz before sending through Wi-Fi. This device is being powered by a 5V 16,000mAh Li-ion battery power supply. The 2 channel signal from the analog frontend was transferred to a computer server for data collection. Lead II data were then stored as a 6-seconds ECG epoch (1536 samples) in order to visualize and classified with a pre-trained CNN. There were two CNN models with a residual network architecture (ResNet) conducted in this study: SmallNet and BigNet. SmallNet classified an ECG signal into 3 classes which are nor-mal, abnormal, and noise, then reported them in a probability form on the displaying screen. BigNet can be used to identify 10 different types of arrhythmia which are normal, left bundle branch block beat, right bundle branch block beat, atrial premature beat, premature ventricular contraction, ventricular escape beat, a fusion of ventricular and normal beat, paced beat, other abnormal beat and noise. This platform was a preliminary step of a bedside ECG monitoring platform that can visualize the ECG signal with a probability of normal, abnormal and noise with an accuracy of 99.1%, which can be adapted for a reduction of the false alarm rate. Moreover, the stored data can also call back later to determine the type of arrhythmia with an ac-curacy of 98.5% if any abnormality occurred.