Algorithms for Advancing the Patient-Personalization of Preoperative Planning in Image-Guided Robotic Surgery

Abstract

This dissertation improves the patient personalization of preoperative planning for a wide variety of surgical applications. These improvements are accomplished via the creation of several novel algorithms in two general areas: (1) tool path planning for autonomous bone milling, and (2) creating statistically valid safety margins to protect critical structures from damage. While bone milling is a common application of surgical robotics, most published tool path planners are 2.5D, layer by layer, planners. Chapter 2, presents a voxelized framework that can implement a variety of milling strategies. A novel true 3D path planner is then presented. This path planner is tested in simulation and via milling mastoidectomies in temporal bone specimens. The 3D planner generates shorter tool paths than a standard 2.5D path planner for all but the smallest tested volume. The efficiency gain increases linearly with the volume of the milled cavity. This algorithm also gives the surgeon more control over the tool’s path for an individual patient. Chapter 3 creates spatially varying safety margins surrounding critical structures identified by the surgeon. This safety margin is generated using statistical models of uncertainty in fiducial point-based registration. These margins are typically set heuristically with uniform thickness, which does not reflect the anisotropy and spatial variance of registration error. The algorithm was applied to five CT scans with planned mastoidectomies. Safety margins were generated that satisfied the specified safety levels in every case. The algorithm presented in Chapter 3 requires rigid point-based fiducial registration. Chapter 4 presents a novel algorithm that removes this restriction and is applicable to any registration modality. The algorithm was used to generate safety margins surrounding kidney tumors in partial nephrectomy. This margin is taken to prevent positive margins (i.e.\ tumor tissue accidentally left behind). Long term patient outcomes are best when the most renal tissue is preserved. This desire to remove as little tissue as possible, while fully removing the tumor, makes partial nephrectomy an excellent procedure for the algorithm. Spatially varying safety margins were created for four tumors and in each case they were smaller than the constant thickness margins developed for the same tumors.