Point Cloud Segmentation of Breast Ultrasound Regions to be Scanned by Fusing 2D Image Instance Segmentation and Keypoint Detection

Automatic segmentation of the region of interest to be scanned (ROIS) is one of the key tasks to achieve autonomous scanning of ultrasound robots. By analyzing breast ROIS in an unstructured environment, a point cloud segmentation framework for the breast ultrasound ROIS that incorporates 2D image instance segmentation and keypoint detection is proposed. Firstly, 2D image instance segmentation based on the improved SOLOv2 is performed to extract the human torso point cloud in the unstructured environment. Then, based on YOLO-Pose, the keypoints of human body are automatically detected, and the upper and lower lines of breast ROIS are obtained according to the constraint matching relationship between breast ROIS and keypoints, and finally the point cloud of breast ROIS is obtained effectively and accurately. Experiments show that the proposed framework can achieve automatic segmentation of breast ROIS within 2 s.