Investigation of Interactive Segmentation for Bifurcation of Carotid Artery on 3D Ultrasound Image Volume

Ultrasound (US) imaging is widely used to diagnose carotid atherosclerosis (CA) caused by carotid plaque since it is non-invasive and economic. Current automatic plaque segmentation method based on deep learning shows high accuracy on common carotid artery but poor performance at the bifurcation of carotid artery. This study proposes an interactive segmentation algorithm that accurately segments carotid plaque at the bifurcation using limited user interaction. The algorithm utilizes a 3D ultrasound imaging system and a 3D U-net segmentation network to obtain masks of media-adventitia boundary (MAB) and lumen-intima boundary (LIB). The results demonstrate improved accuracy over conventional automatic segmentation methods with DSC /HD95 values of 0.952/0.964 for MAB and 0.937/1.068 for LIB, respectively. The proposed algorithm shows potential of clinical implications for the diagnosis of carotid atherosclerosis at bifurcation.