MACHINE LEARNING BASED HEART ANALYSIS IN REAL TIME, USING LOW-COST ECG SENSOR

Abstract

This work explores the use of an ESP32 microcontroller, equipped with AD8232 ECG sensor, to analyze ECG patterns and detect heart abnormalities in real-time, with the help of machine learning model. Data is collected using low-cost, self-assembled ECG system, focusing on ECG signals. Machine learning model, specifically the hybrid CNN model, is employed for feature extraction and temporal analysis, allowing for real-time detection of various heart abnormalities. Evaluation of the models is done through metrics like accuracy, sensitivity, F1 score, and AUC. Practical experiments are to be conducted using a self-assembled system. Special software using open-source tools is implemented, to bridge the hardware and software. Total cost of the self-assembled system reached 15 USD. System was used to classify 6 different heart abnormalities, namely I degree atrioventricular block (AV), Atrial Fibrillation (AF), Sinus Tachicardia (ST), Sinus Bradycardia (SB), Right Bundle Branch Block (RBBB), Left Bundle Branch Block (LBBB). The machine learning model was tested using LUED database, a public databases from the Physio.net platform. Model reached performances of over 0.9 AUC for all of the 6 classes, and F1 score of 0.8+ for AF and RBBB & LBBB classes. The web application, including the backend server, were implemented and deployed using open-source tools and free platforms. The web application is able to visualize the real-time ECG readings and additional metric with almost zero latency. Overall, this work shows the potential of low-cost wearable devices, combined with advanced machine learning techniques, to analyze heart health and to provide real-time monitoring for ECG patterns, being a promising integration in healthcare.