Carotid artery image segmentation

Abstract

The main process causing most cardiovascular diseases is atherosclerosis, which is responsible for the thickening of the major arteries walls. Concretely, the intimamedia thickness (IMT) of the carotid artery wall is an early and effective marker of atherosclerosis progression. The measurement of the IMT is directly extracted from the segmentation of two different layers of the carotid artery wall. In this project, we present three fully automated techniques to perform the segmentation of these two layers of the carotid artery wall using B-mode ultrasound images. The segmentation of the carotid artery wall is a challenging problem due to image noise, artefacts and image shape, intensity and resolution variability. One of the developed methods is based on lumen detection. It first detects the lumen region of the carotid artery and then it seeks the both layers using the differences between the intensity values of the image. The other two methods are based on a classification system, considering the image segmentation problem as a classification problem of the image pixels into interior or exterior of the region formed by the two layers. One of them uses the random forest classifier and the other one uses the stacked sequential learning scheme with random forest as a base learner. We validate the proposed techniques using a data set of B-mode images obtained from a clinical institution and we compare its performances.