Automatic Image-based Localization and Modelling for Laparoscopy (Automatische beeldgebaseerde localisatie en modellering voor laparoscopie)

Publication date: 2013-04-02

Author:

Moll, Markus
Van Gool, Luc

Keywords:

PSI_VISICS

Abstract:

Keyhole surgery has become standard procedure in modern surgery. Theminimal invasiveness offers great advantages for the patient’s recovery, limits unsightly scarring, and reduces post-surgical complication risks. However, there are severe challenges for surgeons. The surgical site can only be viewed through an endoscope. This affects the surgeon’s depth perception as well as his hand-eye coordination. Robot-assisted surgery is increasingly popular, in particular in conjunction with minimally invasive surgery. In this thesis, we investigate the use of monocular rigid and nonrigid3D reconstruction methods in laparoscopy, i.e. endoscopic surgery in theabdominopelvic cavity. To this end, we analyze and improve on state of theart methods for rigid and nonrigid reconstruction and their use in laparoscopy. We build a real-time rigid reconstruction system that is useful on its own and furthermore serves as a starting point for nonrigid deformation tracking. For nonrigid reconstruction, local methods are more suitable in the laparoscopic setting, as the deformations are complex on a global scale. We look into the feasibility of these nonrigid reconstruction methods and lift some of their limitations with respect to laparoscopic surgery.Intra-operative 3D reconstruction has a number of applications in laparoscopy, and the use of monocular reconstruction techniques allows us to immediately apply our methods in clinical practice, where predominantly monocular endoscopes are used. Applications include the use of augmented reality tools to aid the surgeon, simplified automation of robotic surgery by providing robots with an optical sensory system, and use for tele-surgery purposes where surgeons at a remote site can have a 3D view of the surgical site.