ABSTRACT
Three-dimensional ultrasound (3DUS) provides important plaque volume measurements for the assessment of carotid plaque burden and monitoring changes in carotid atherosclerosis in response to medical treatment. Carotid 3DUS-based measurements require the segmentation of the media-adventitia boundary (MAB), lumen-intima boundary (LIB), and plaque boundaries from the carotid 3DUS images. However, manual analyses are time-consuming and depend on the experience of the examiners. In this chapter, we introduce recent 3DUS-based carotid plaque segmentation methods and their performance. Furthermore, we focus on the deep-learning-based methods for plaque segmentation, including the model architectures, loss functions, and post-processing methods. For the model architecture, U-Net is the most popular architecture in medical image segmentation, U-Net++ is an improvement of U-Net, and Voxel-FCN is a 3D convolution network for MAB and LIB segmentation from carotid 3DUS images. For the loss function design, cross-entropy and DSC loss are two popular loss functions of medical image segmentation. We further describe the details of weighted binary-entropy loss and the triple DSC loss to improve the segmentation performance. For post-processing methods, the multi-class continuous max-flow algorithm and the multiple CNN ensemble algorithm can be used to refine the deep-learning segmentations.
