Label-based Optimization of Dense Disparity Estimation for Robotic Single Incision Abdominal Surgery

Minimally invasive surgical techniques have led to novel approaches such as Single Incision Laparoscopic Surgery (SILS), which allows the reduction of post-operative infections and patient recovery time, improving surgical outcomes. However, the new techniques pose also new challenges to surgeons: during SILS, visualization of the surgical field is limited by the endoscope field of view, and the access to the target area is limited by the fact that instruments have to be inserted through a single port. In this context, intra-operative navigation and augmented reality based on pre-operative images have the potential to enhance SILS procedures by providing the information necessary to increase the intervention accuracy and safety. Problems arise when structures of interest change their pose or deform with respect to pre-operative planning, as usually happens in soft tissue abdominal surgery. This requires online estimation of the deformations to correct the pre-operative plan, which can be done, for example, through methods of depth estimation from stereo endoscopic images (3D reconstruction). The denser the reconstruction, the more accurate the deformation identification can be. This work presents an algorithm for 3D reconstruction of soft tissue, focusing on the refinement of the disparity map in order to obtain an accurate and dense point map. This algorithm is part of an assistive system for intra-operative guidance and safety supervision for robotic abdominal SILS . Results show that comparing our method with state-of-the-art CPU implementations, the percentage of valid pixel obtained with our method is 24% higher while providing comparable accuracy. Future research will focus on the development of a real-time implementation of the proposed algorithm, potentially based on a hybrid CPU-GPU processing framework.