Chapter 6 - Binary descriptor design for the automatic detection of coronary arteries using metaheuristics

Abstract

The automatic binary descriptor design for vessel enhancement using morphological operators is a O(2n) computational complexity problem. Consequently, this problem has been mainly addressed by an empirical process and the high-dimensional search space has not been explored properly. In this chapter, the automatic binary descriptor design has been addressed by comparing three different metaheuristics for the coronary artery segmentation problem in x-ray angiograms. The main advantage of the automatic binary descriptor design is that the search space is explored properly avoiding an empirical process or a priori knowledge by an expert to design deterministic descriptors. In the vessel enhancement step, the binary descriptor is designed using metaheuristics to be applied with the morphological top-hat operator. In this step, three metaheuristics are compared with five vessel detection methods using as an evaluation metric the area (Az) under the receiver operating characteristic curve. From computational experiments, the metaheuristic of iterated local search (ILS) achieved an Az value of 0.9682 using a training set of 50 angiograms and 0.9614 using a test set of 50 angiograms. Moreover, in the second step, the resulting top-hat filter response is binary classified as vessel and nonvessel pixels by comparing five state-of-the-art thresholding methods using the accuracy measure as an evaluation metric. Finally, the comparative analysis reveals that the top-hat operator based on the ILS for vessel detection and the intraclass variance thresholding method obtained the high segmentation accuracy of 0.9621 using the test set of 50 angiograms.