DEEP LEARNING–DRIVEN PATIENT-LEVEL CAD ANALYSIS IN MRI

Abstract

 Early diagnosis of coronary artery disease (CAD) is critical for improving patient outcomes. This work presents a patient-level deep learning pipeline for automated CAD detection from cardiac MRI, designed to distinguish Normal and Sick cases under realistic clinical conditions. The framework processes patient-grouped studies with standardized pre-processing and employs a balanced data generator with light augmentations to address class imbalance and imaging variability. A convolutional neural network is trained using strict patient-wise splits to prevent data leakage and ensure robust generalization. Comprehensive quantitative and visual evaluation supports interpretability and model refinement. This work investigates the role and contribution of deep learning, especially fully convolutional networks and convolutional neural networks, toward the improvement of accuracy and automation in cardiac MRI analysis.