Segmentation and labelling of intra-operative laparoscopic images using structure from point cloud

We present in this paper an automatic method for segmenting and labelling the liver and its surrounding tissues in intraoperative laparoscopic images. The goal is to distinguish between the different structures that compose a intra-operative hepatic surgery scene to improve common registration tasks. Our segmentation method considers the scene as a 3D structured point cloud to exploit powerful informations such as curvature and normals, in addition to visual cues that permit to efficiently classify the scene. Experiments performed on challenging human hepatic surgery confirm that accurate segmentation and labelling are possible using our approach.