Segmentation-Based CCTA Image Analysis to Support the Diagnosis of Coronary Artery Disease

A early diagnosis of a coronary artery disease (CAD) can reduce the risk of serious clinical events, to this aim it is important preliminary evaluate risk index. Agatston score and calcium volume score are among the main indicators used to predict the presence of calcium in the artery coronary. In this paper we focused on the implementation of an automatic procedure able to predict an accurate estimation of these diagnostic parameters. The method is based on U-Net convolutional neural network architecture built from residual units in order to segment the artery calcification. To improve the performance of the model a region growing algorithm is applied to the seed obtained by U-Net. A comparison between the estimations of the Agatston score and the calcium volume score obtained with and without region growing highlights as the performance of the model increases the accuracy at around 97% of the segmentation with region growing.