Cardiac Deep Learning Model for Congenital Heart Disease Recognition

Congenital heart disease (CHD) is a worldwide common congenital disability, and according to a recent WHO report, more than 60,000 babies are born with CHD yearly. Furthermore, total death rates due to CHD reached 16.49% of the total baby deaths. CHD has thirty-five types with various abnormalities in the heart, including abnormal, incomplete, or some parts missing in the heart. Similarly, there are leaky valves and holes in the chamber partitions of the heart. A few types of CHD can be detected during pregnancy and before birth. Patients with CHD need to be diagnosed early, whereas many children die undiagnosed due to CHD because of a lack of awareness about CHD in developing countries. Moreover, in most cases, if CHD is diagnosed in the early stage of life, most people can not afford the cost of CHD surgeries and other procedures for their kids. Recognizing this gap, we propose a deep learning-based model we named the Cardiac Deep Learning Model (CDLM) that can detect this abnormality easily and in the early stage of life with CT-scanned images. We use a segmentation model which segments the four chambers of the heart and then followed by the blood pool stage. We apply a graph-matching method to extract connection information and to know the categories of all the vessels. For experimental work, we use a publically available dataset having 68 CT images of the heart, which covers 14 different kinds of abnormalities. Our proposed method reported better results as compared to the results reported previously.