Automated Segmentation of Coronary Arteries using Attention-Gated UNet for Precise Diagnosis

Abstract

Computed Tomography Angiography (CTA) has revolutionized coronary artery disease diagnosis and treatment with its high-resolution and non-invasive advantages. Precision in coronary artery segmentation is critical for accurate diagnosis and effective treatment. In this study, we introduce an innovative two-stage approach utilizing fully convolutional neural networks (CNNs) for coronary artery segmentation. Our model combines coarse segmentation with fine segmentation, significantly enhancing accuracy. This dual-stage strategy outperforms conventional single-stage methods, as validated through empirical evaluations. Our approach demonstrates a substantial performance improvement, with a MEAN Jaccard Similarity of 0.8217 and a MEAN Dice Similarity Coefficient of 0.9005, affirming its potential in advancing medical imaging diagnostics and improving coronary artery segmentation.