Multi-task Learning Approach for Automatic Diagnosis and Segmentation of Carotid Atherosclerosis Using Portable 3D Ultrasound

Carotid artery segmentation and atherosclerosis identification are important in the diagnosis of carotid atherosclerosis. Conventional 2D ultrasound devices may overlook some plaque information of carotid atherosclerosis due to angle issues. This study proposed a multi-task learning method for automatically diagnosing carotid atherosclerosis and simultaneously segmenting carotid arteries to extract the 3D volume of plaque for further analysis. A 3D U-net was firstly employed for carotid segmentation, followed by integrating the segmentation results and original data volume into the classification module composed by another adapted 3D U-net. The accuracy of atherosclerosis identification was 90.0% with a sensitivity of 91.7% and a specificity of 89.3%. Meanwhile, the DSC from our approach were 0.9189 for lumen-intima boundary (LIB) and 0.9444 for vessel-wall-volume (VWV). The result of the proposed algorithm showed the potential of clinical implications for the diagnosis of carotid atherosclerosis using the portable 3D ultrasound imaging technique.