CADNet: A Lightweight Neural Network for Coronary Artery Disease Classification Using Electrocardiogram Signals.

Abstract

Coronary Artery Disease (CAD) is characterised by a diminished capacity of the coronary arteries to supply sufficient blood, oxygen and nutrients to the heart. It primarily develops due to the presence of fat deposits and arterial plaques, and it is a leading cause of global mortality. Given the limited accessibility, high cost, and inconvenience of invasive diagnostic tools, we propose a lightweight one-dimensional convolutional neural network for CAD classification using non-invasive electrocardiography (ECG) signals. The proposed model, CADNet, consists of two key components: Feature Encoding and Compact Pooling. The feature encoding block extracts key temporal characteristics from ECG data using a convolutional layer, while the compact pooling block reduces temporal resolution, preserving essential ECG features for CAD diagnosis. CADNet comes with a novel data purification process to optimise computational efficiency and maintain high diagnostic accuracy. This approach aids convergence, significantly reduces the model parameters, and improves the model's ability to detect CAD patterns. Our extensive experiments with four diverse datasets show that CADNet achieves an average 99.3% accuracy, with 2,586 trainable parameters, surpassing state-of-the-art models performance.