A Hierarchical Ensemble Deep Learning Framework for Angiographic Blood Vessel Segmentation and Coronary Artery Disease Classification

Abstract

Artificial Intelligence (AI) and deep learning have revolutionized medical image analysis by enabling accurate, efficient, and automated disease diagnosis across diverse healthcare applications. In cardiovascular medicine, these technologies have significantly improved the analysis of coronary angiographic images, supporting early detection and clinical decision-making for coronary artery disease (CAD). However, accurate angiographic blood vessel segmentation and disease classification remain challenging due to low image contrast, brightness inhomogeneity, imaging noise, and the limited feature representation capability of single deep learning models. To address these challenges, this paper proposes a hierarchical ensemble-based deep learning framework for angiographic blood vessel segmentation and coronary artery disease classification. First, Principal Component Analysis (PCA), Contrast Limited Adaptive Histogram Equalization (CLAHE), and the Toggle Contrast Operator (TCO) are employed to enhance angiographic image quality. Secondly, an adaptive Gaussian kernel probability density function (PDF)-based matched filtering technique, followed by entropy-based thresholding, length filtering, and masking, is applied for accurate coronary vessel segmentation. Thirdly, complementary deep features are extracted using EfficientNet-B0, VGG-16, and ResNet-152 and integrated through an ensemble feature fusion strategy. Finally, the fused features are classified using a Softmax classifier. Experimental results demonstrate that the proposed framework achieves an accuracy of 91.23%, precision of 91.84%, sensitivity of 90.97%, specificity of 93.41%, F1-score of 91.40%, MCC of 0.825, Cohen's Kappa of 0.823, and an AUC of 0.951. Comparative evaluation against conventional machine learning and state-of-the-art deep learning models demonstrates that the proposed framework provides superior classification performance, robustness, and diagnostic reliability, "making it a promising computer-aided clinical decision support tool" l for coronary artery disease diagnosis.