V. Castaneda, S. Atasoy, D. Mateus, N. Navab, A. Meining Reconstructing the Esophagus Surface from Endoscopic Image Sequences 5th Russian-Bavarian Conference on Bio-Medical Engineering, Munich, Germany, July 1-4 2009. (bib)
	The current screening and surveillance protocol for early stage esophageal cancer detection consists of acquiring biopsies under endoscopic guidance. We are interested in assisting physicians in tracking the evolution of the tissue by using the information provided by the endoscopic videos. In particular, the current protocol lacks a procedure for guiding the physician by retargeting the same biopsy locations in the follow-up examinations. In this work, we explore the possibility of creating a 3-dimensional (3D) patient-specific visualization of the esophagus surface from the endoscopic image sequences. One of our major goals is to map the previous biopsy sites on the 3D visualization of the tissue surface and assist the endoscopist in retargeting the same locations. However, for the reconstruction of the esophagus surface from endoscopic videos several challenges need to be addressed. The esophagus tissue presents big deformations which can greatly modify appearance of the tissue. Moreover, several factors affect the quality of the images, e.g. liquid inside the esophagus generates specularities and fast motion of the camera leads to blurred images. In this work, we propose to model the reconstruction problem using the simultaneous localization and mapping (SLAM) approach. In particular, we would like to demonstrate the feasibility of monoSLAM (Monocular SLAM) algorithm to an endoscopic image sequence to get a first estimation of the 3D surface and its appearance. In brief, the goal of the SLAM formulation is to simultaneously estimate the 3D position of points on the surface of the esophagus (mapping) and the endoscopic camera position (localization). The monocular SLAM algorithm solves the problem by tracking groups of feature points in the 2D image sequence. For the robustness of the feature tracking we propose the use of Scale-invariant features (SIFT) applied directly on endoscopic images to detect and describe the features points. Our preliminary results demonstrate that SLAM is a promising tool for 3D tissue visualization and mapping. However, due to the difficulties of in-vivo sequences only small regions of the surface can be reconstructed. Currently, we analyze the difficulties and we plan to introduce further modifications to the algorithm to allow for its in-vivo application.
	This material is presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each authors copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.