Learning to Focus on Boundary Information forMedical Image Segmentation

Abstract

Boundary detection is a challenging problem in medical image segmentation. Clear boundaries yield a good segmentation result. In this paper, firstly, we design a boundary segmentation network for the detection and segmentation of brain tissues. Secondly, to help distinguishing between the boundaries for the three different brain tissues, we design the boundary information module (BIM ). Then, we added a boundary attention gate (BAG) to each end layer of the encoder to capture more local details. Extensive experiments verified the high performance of our proposed model for the detection and segmentation of brain tissues.