MRI is a promising noninvasive technique for characterizing atherosclerotic plaque composition in vivo, with an end-goal of assessing plaque vulnerability. Because of limitations arising from acquisition time, achievable resolution, contrast-to-noise ratio, patient motion, and the effects of blood flow, automatically identifying plaque composition remains a challenging task in vivo. In this article, a segmentation method using maximum a posteriori probability Bayesian theory is presented that divides axial, multi-contrast-weighted images into regions of necrotic core, calcification, loose matrix, and fibrous tissue. Key advantages of the method are that it utilizes morphologic information, such as local wall thickness, and coupled active contours to limit the impact from noise and artifacts associated with in vivo imaging. In experiments involving 142 sets of multi-contrast images from 26 subjects undergoing carotid endarterectomy, segmented areas of each of these tissues per slice agreed with histologically confirmed areas with correlations (R(2)) of 0.78, 0.83, 0.41, and 0.82, respectively. In comparison, manually identifying areas blinded to histology yielded correlations of 0.71, 0.76, 0.33, and 0.78, respectively. These results show that in vivo automatic segmentation of carotid MRI is feasible and comparable to or possibly more accurate than manual review for quantifying plaque composition.
Magn Reson Med, 2006. (c) 2006 Wiley-Liss, Inc.