Non-linear-Optimization Using SQP for 3D Deformable Prostate Model Pose Estimation in Minimally Invasive Surgery

Augmented Reality began to be used in the last decade to guide and assist the surgeon during minimally invasive surgery. In many AR-based surgical navigation systems, a patient-specific 3D model of the surgical procedure target organ is generated from preoperative images and overlaid on the real views of the surgical field. We are currently developing an AR-based navigation system to support robot-assisted radical prostatectomy (AR-RARP) and in this paper we address the registration and localization challenge of the 3D prostate model during the procedure, evaluating the performances of a Successive Quadratic Programming (SQP) non-linear optimization technique used to align the coordinates of a deformable 3D model to those of the surgical environment. We compared SQP results in solving the 3D pose problem with those provided by the Matlab Computer Vision Toolkit perspective-three-point algorithm, highlighting the differences between the two approaches.