Exploiting Vector Attention and Context Prior for Ultrasound Image Segmentation

Lu Xu, Shengbo Gao, Lijuan Shi, Boxuan Wei, Xiaowei Liu, Jicong Zhang, Yihua He

Abstract: Highlights•Vector self-attention layer achieves spatial and channels reasoning synchronously.•Context aggregation loss considers global structure and contextual prior.•Contextual prior focuses on context dependencies of intra-classes and inter-classes.•The model achieves the best performance on two ultrasound segmentation datasets.