A Novel Evolutionary Morphological Descriptor for the Segmentation of Coronary Arteries

This paper presents a novel evolutionary morphological descriptor for the detection of coronary arteries in X-ray angiograms. The method consists of the vessel detection and segmentation steps. In the first step, the univariate marginal distribution algorithm (UMDA) is used to design the optimal binary template for detecting vessel-like structures by applying the morphological top-hat operator in the spatial image domain. The results of the evolutionary descriptor are compared with those obtained by using methods based on the Hessian matrix, Gaussian filters and top-hat operator. The proposed method obtained the highest performance in terms of the area (A_z) under the ROC curve (A_z = 0.9661) compared with five state-of-the-art vessel detection methods using a training set of 50 angiograms and A_z = 0.9532 with an independent test set of 50 angiograms. In the second step, six thresholding methods are compared in terms of segmentation accuracy in order to classify vessel and nonvessel pixels from the evolutionary filter response. Finally, the experimental results for vessel segmentation with the proposed method provided an accuracy of 0.9580 with the test set of angiograms.