Appearance Modelling and Reconstruction for Navigation in Minimally Invasive Surgery

Abstract

Minimally invasive surgery is playing an increasingly important role for patient care. Whilst its direct patient benefit in terms of reduced trauma, improved recovery and shortened hospitalisation has been well established, there is a sustained need for improved training of the existing procedures and the development of new smart instruments to tackle the issue of visualisation, ergonomic control, haptic and tactile feedback. For endoscopic intervention, the small field of view in the presence of a complex anatomy can easily introduce disorientation to the operator as the tortuous access pathway is not always easy to predict and control with standard endoscopes. Effective training through simulation devices, based on either virtual reality or mixed-reality simulators, can help to improve the spatial awareness, consistency and safety of these procedures. This thesis examines the use of endoscopic videos for both simulation and navigation purposes. More specifically, it addresses the challenging problem of how to build high-fidelity subject-specific simulation environments for improved training and skills assessment. Issues related to mesh parameterisation and texture blending are investigated. With the maturity of computer vision in terms of both 3D shape reconstruction and localisation and mapping, vision-based techniques have enjoyed significant interest in recent years for surgical navigation. The thesis also tackles the problem of how to use vision-based techniques for providing a detailed 3D map and dynamically expanded field of view to improve spatial awareness and avoid operator disorientation. The key advantage of this approach is that it does not require additional hardware, and thus introduces minimal interference to the existing surgical workflow. The derived 3D map can be effectively integrated with pre-operative data, allowing both global and local 3D navigation by taking into account tissue structural and appearance changes. Both simulation and laboratory-based experiments are conducted throughout this research to assess the practical value of the method proposed.