Automated Coronary Artery Disease Severity Grading Using Deep Learning

Abstract

Coronary Artery Disease (CAD) is a common cardiovascular disease worldwide and requires accurate and early detection to avoid serious complications such as heart attacks and cardiac arrests. Invasive Coronary Angiography (ICA) remains the most reliable method for assessing the degree of arterial stenosis; however, its interpretation is often subjective and prone to human error. This project aims to overcome these limitations by utilizing a deep learning-based framework for automated CAD classification using ICA images. The classification of image patches into lesion and non-lesion categories across different lesion severity ranges is performed using five Convolutional Neural Network (CNN) architectures: DenseNet, ResNet, LeNet, AlexNet, and VGGNet, after implementation and evaluation. High-quality model training is achieved by preprocessing the dataset through vessel segmentation, patch extraction, class balancing, and data augmentation. Measurement of performance is done by using accuracy, precision, recall, F-measure, and area under the ROC curve (AUC). The experimental results demonstrate that DenseNet and ResNet perform better, with an F-measure that exceeds 90% and an AUC that exceeds 98%, effectively identifying severe lesion regions. By using the proposed framework, CAD diagnosis can be improved through the use of a robust, automated, and clinically applicable solution that improves diagnostic precision and reduces manual workload for cardiologists.