An auxiliary learning network for carotid ultrasound image classification

Carotid plaque classification tasks perform an important part in preventing the occurrence of ischemic strokes. Carotid ultrasound images provide a clear picture of the location and status of plaque which helps the professional doctor in the diagnosis of carotid atherosclerosis. However, existing methods ignore the relevance of different carotid plaque tasks. To exploit the different features of carotid ultrasound images, we propose an auxiliary learning framework. The framework uses carotid plaque classification as the primary task and carotid plaque segmentation as an auxiliary task. The framework uses hard parameter sharing to learn the different features of carotid ultrasound images. The results show that the proposed method is beneficial for improving the carotid plaque classification performance.