Patch based 3D reconstruction of the liver surface from laparoscopic videos

Abstract

Endoscope is commonly used in Laparoscopic surgery. It allows to the doctor to perform the surgery without opening a cavity. Therefore, it allows the surgeon to perform smaller incisions and consequently diminish the risk of infection. An investigation regarding the reconstruction of the liver surface from a sequence of images obtained from an endoscope video has been done by the Vision Institute of TUHH, in cooperation with the Medical Center Hamburg-Eppendorf. The goal of this paper is to optimize that reconstruction. Previous works in the institute allowed, using matlab functions, to obtain a first reconstruction. Having the matrices of the transformations between cameras and the reconstructed points in 3D, the goal is to optimize that using a patch based 3D reconstruction. The method consists of the construction of a patch for every point and trying to find a good normal for the patch. Some tools are used for the optimization like normal cross correlation that is the similarity function used for obtaining the match of the corresponded 2D windows in different cameras, and a nonlinear optimization method to look for the best normal who gives the best match between the windows. The main objective is to evaluate this method and draw conclusions about its utility and viability: can this method be used for the optimization in the case of the endoscope video of a liver surface.