Segmentation of Coronary Arteries from CTA axial slices using Deep Learning techniques

Abstract

Coronary Artery disease (CAD) is a type of Cardio Vascular disease caused due to disorders in blood vessels of the heart. Stenosis is sudden narrowing or blockage in coronary arteries and this happens because of cholesterol formation, fatty deposition, and damages in blood cells. In order to detect stenosis in Computed Tomography Angiography (CTA) images, the segmentation of the coronary artery is essential. In this paper, we propose a deep learning based model to segment the coronary arteries from 2D slices of CTA heart images and reconstruct them into 3D coronary artery. The coronary arteries may not present in all the 2D slices, so a combination of Convolutional Neural Network and Recurrent Neural Network (CNN-RNN) model is used to identify the presence of coronary arteries in the 2D slices. The identified coronary arteries are then segmented using the U-Net model. The segmented coronary arteries are then reconstructed into 3D images using the Maximum Intensity Projection (MIP) reconstruction algorithm in order to analyze the presence of stenosis. The proposed system was evaluated for segmentation using IOU (Intersection Over Union) and for reconstruction using Structure Similarity Index metric (SSIM) for the CTA images obtained from Bilroth hospitals, Chennai.