Automatic GrabCut based lung extraction from endoscopic images with an initial boundary

Endoscopic images provide doctors with valuable information in both diagnosis and surgery. In a thoracoscopic surgery, locating lung part based on endoscopic images is difficult since lungs vary sharply according to respiration. In this case, correctly extracting lung part plays a crucial role in intraoperative navigation. Although many efficient image segmentation approaches have been developed in the last two decades, they can rarely achieve reliable performance in lung segmentation due to the high deformability of lungs, similarity between lungs and background, unstable movement of the endoscope, and dynamic appearance changes of both lungs and background. In this research, we propose an effective approach for extracting lungs during endoscopic videos. The proposed algorithm is based on GrabCut which derives from max-flow min-cut theorem. However, unlike GrabCut that needs user interaction for each frame segmentation, the proposed method only requires an initial boundary of the first frame. Furthermore, it utilizes motion and boundary information to facilitate GrabCut to achieve a global optimum. The robustness of the proposed approach has been validated by experiments using clinical lung endoscopic videos.