Auto-Detection and Classification of Arrhythmias with ECG Signal Analysis

Overview

Cardiac arrhythmias represent a significant global health challenge, affecting millions annually and frequently leading to misdiagnosis and inappropriate treatment. Consequently, 75% of arrhythmia patients suffer from progressively worsening heart conditions The complexity of arrhythmias, along with the subtleties that distinguish one type from another, makes accurate diagnosis crucial yet challenging, especially for general practitioners and less experienced cardiologists. Studies indicate that the rate of correct diagnosis by non-specialists stands at a low of 41% and by specialists at 78%, leading to delayed or incorrect treatments. My goal was to develop a novel algorithm using convolutional neural networks (CNNs) to automate the detection and classification of cardiac arrhythmias with accuracy that matches or exceeds that of board-certified cardiologists. By utilizing a dataset from a comprehensive ECG database, we trained four distinct CNN architectures on over 10,000 ECG recordings, covering several common arrhythmias, including atrial fibrillation, ventricular tachycardia, and bradycardia. The most effective model, ResNet-50, demonstrated exceptional performance in our end-to-end algorithm, achieving a classification accuracy of 94.2% across eight arrhythmia types. This model presents a promising tool for enhancing the accuracy of arrhythmia diagnosis, potentially improving patient outcomes by facilitating timely and accurate treatment decisions for a broad spectrum of cardiac arrhythmias