Improving image-guided preoperative planning in deep brain stimulation procedures

Abstract

Deep brain stimulation is a stereotactic surgery widely used to treat neurological disorders such as Parkinson’s disease. This procedure involves inserting a permanent electrode into the deep brain and sending electrical pulses to stimulate target nuclei. To achieve maximum therapeutic benefits, meticulous planning is required prior to the surgery to identify a safe insertion trajectory and a precise stimulation target point in patient images. These are two challenging tasks because of the complicated factors affecting the implantation, the insufficient image quality, the small-sized nuclei structures, and the millimetric accuracy requirement. In this dissertation, we present a set of innovative and efficient image processing methods to provide surgical guidance in preoperative planning. For trajectory planning, we develop a system that optimizes the costs of candidate trajectories derived from multiple surgical constraints. For target localization, several studies are conducted to (1) localize visible anatomical landmarks and planes for indirect targeting, (2) evaluate statistical atlas-based targeting approaches using various nonrigid image registration algorithms, (3) learn the target position from past patients with a multivariate regression model, and (4) localize target structures using a priori shape information derived from manual delineations in high-field images. State-of-the-art learning-based, atlas-based, and shape-based techniques are investigated and adapted in these studies to achieve the desired accuracy and robustness despite limited contrast, imaging artifacts, and low spatial resolution. Large-scale translational experiments are performed at multiple institutions to demonstrate the efficacy of the presented techniques to support clinical decisions.