作者: Zeynettin Akkus , Nico De Jong , Antonius FW Van Der Steen , Johan G Bosch , Stijn CH Van Den Oord
关键词:
摘要: Segmentation of carotid plaques in standard B-mode ultrasound is challenging due to irregular lumen shapes, noise, artifacts, and plaque echolucency. To overcome these challenges, we propose a novel segmentation method which exploits the benefits simultaneously acquired (BMUS) contrast enhanced (CEUS). We first estimate anatomical motion from BMUS image sequence, using nonrigid intensity-based registration, apply CEUS sequences for compensation. average motion-compensated obtain single BMUS&CEUS images with improved signal-to-noise ratio, serve as ‘epitome’ images. vessel detection distinguish multiple branches epitome. The lumen-intima layer segmented epitomes by joint-histogram classification approach, followed 1D dynamic programming procedure. Then, media-adventitia dual (2D) (DDP). As adventitial wall are almost parallel each other, DDP used find two smooth lines combining their costs. For validation, layers 13 arteries atherosclerotic were manually observers compared automated results. manual segmentations was considered ground-truth. layer, ± std. dev. root-mean-square-error 283±123µm. inter-observer variability 261±128µm. 334±170µm. 225±130µm. differences between ground-truth contours same order those observers. In conclusion, present an accurate, robust, fully combined best our knowledge, this study exploit combination solve challenges plaques.