Stereo vision-guided laser microsurgery

Abstract

Transoral laser microsurgery (TLM) provides the most advanced microscopic technique for contact-less, atraumatic treatment of oral, pharyngeal, and in especially laryngeal carcinomas. The primary goal of the laser ablation is radical removal of the lesion. This necessitates the selection of sufficiently large resection margins and thus conflicts with a further principle of surgery which is function preservation. Preserving as much healthy tissue as possible requires very accurate laser positioning that, as of today, demands a highly experienced surgeon. Even though the advantages of TLM are manifold, there are substantial technical limitations compromising the post-operative outcome and hence the quality of life of the patient. Major challenges do not solely arise from a limited field of view and range of motion, but in particular from inaccurate laser focusing, inadequate imaging of the submucosal extent, suboptimal incision planning and loss of ablation accuracy due to soft tissue motion. To overcome these limitations, this dissertation presents a novel approach for stereo vision-guided laser microsurgery with special emphasis on laryngeal interventions. A variety of laser technologies and computer vision algorithms exist; however, holistic integration of real-time stereoscopic image processing into soft tissue laser surgery, considering the laser characteristics on the one hand and surgical scene information, e.g., structure and non-rigid motion, on the other hand, has not been addressed in its entity thus far. A computational method for stereo vision-based real-time surface estimation is developed as a prerequisite to multimodal registration. The reconstruction accuracy is assessed on an in vivo dataset and a variety of stereo imaging devices considered for microsurgery. A method for laser-to-camera registration is proposed facilitating distance-based laser focus adjustment and thus optimal ablation. To assist the manual focus repositioning process, color-encoded distance superimposed to the live view is implemented as part of the surgeon interface. A further contribution aims at the fusion of optical coherence tomography imaging with stereo vision. A registration and segmentation framework enables the detection and visualization of submucosal changes of laryngeal tissue. Moreover, three-dimensional surface information and laser-to-camera registration are integrated into a stylus-tablet-based interface. Several path planning strategies are evaluated in terms of accuracy and usability. Finally, laser ablation in a dynamic soft tissue environment is addressed with an algorithm for non-rigid tracking. Experiments demonstrate the benefit of live view stabilization during incision planning and closed loop control for motion compensation during laser ablation.