Style Transfer based Coronary Artery Segmentation in X-ray Angiogram

TL;DR

This paper introduces a novel deep learning method combining style transfer and advanced neural network modules for accurate and fast coronary artery segmentation in X-ray angiograms, aiding in coronary disorder diagnosis.

Contribution

It presents a new style transfer-based segmentation technique integrating adaptive instance normalization with dense extreme inception and attention modules, improving accuracy and speed.

Findings

Segmentation accuracy of 0.9658 achieved

Dice coefficient of 0.71 obtained

Method trained on natural images performs efficiently

Abstract

X-ray coronary angiography (XCA) is a principal approach employed for identifying coronary disorders. Deep learning-based networks have recently shown tremendous promise in the diagnosis of coronary disorder from XCA scans. A deep learning-based edge adaptive instance normalization style transfer technique for segmenting the coronary arteries, is presented in this paper. The proposed technique combines adaptive instance normalization style transfer with the dense extreme inception network and convolution block attention module to get the best artery segmentation performance. We tested the proposed method on two publicly available XCA datasets, and achieved a segmentation accuracy of 0.9658 and Dice coefficient of 0.71. We believe that the proposed method shows that the prediction can be completed in the fastest time with training on the natural images, and can be reliably used to diagnose and detect coronary disorders.