Development of a Neural Network-Based CAD system for automated classification of ECG anomalies using wearable technology

In recent times, there has been growing attention towards Computer-Aided Diagnosis (CAD) systems dedicated to the automated analysis of electrocardiograms (ECGs). These systems have gained significant interest due to their potential to facilitate the diagnostic process and improve the precision of cardiac anomaly identification and classification. Furthermore, CAD systems can be situated within the broader context of telemedicine, specifically telecardiology and remote monitoring of cardiac conditions. It is within this context that this thesis work is situated, presenting a method that employs Artificial Intelligence (AI) for the automatic classification of cardiac anomalies. It is based on the electrocardiographic traces (ECG) acquired through a three-lead ECG wearable patch device, commercialised by CGM and designed in collaboration with STMicroelectronics, called 'HI ECG 3-LEAD'. This battery-powered device is designed to be applied with adhesive electrodes to the patient's chest, allowing the acquisition, recording, and transmission of one to three channels of ECG and other physiological parameters (body position, subject's activity status, MEMS data) to an external device via Bluetooth technology. Due to its simplicity, safety, wireless connectivity, and battery life in excess of 24 hours, it is well-suited for monitoring cardiac arrhythmias in non-hospital environments. Additionally, the proposed system aims to provide valuable assistance to cardiologists in the analysis and reporting of ECG signals, reducing the analysis time and increasing the accuracy in identifying and classifying cardiac anomalies. The method consists of two main stages. In the first stage, following appropriate pre-processing of the ECG signal, the positioning of the device on the user's chest is identified to correct any potential electrode misplacement that could compromise the subsequent analysis of cardiac anomalies. This prediction has been addressed as a deep learning problem, specifically as a ternary classification, considering that the device user manual indicates three possible chest electrode placements. In the second stage, a convolutional neural network is employed for detecting and classifying cardiac anomalies. The ECG signals used for training the network were obtained from publicly available arrhythmia databases, as currently, no electrocardiographic traces are available with the CMG Hi 3-Lead device and properly interpreted by cardiologists.