Factors Affecting the Predictive Ability of Computational Models of Subthalamic Deep Brain Stimulation

• Main Author: Bower, Kelsey
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2022
• Subjects: Biomedical engineering
• ISBN: 9798351440422

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) provides striking symptom relief for patients with advanced Parkinson?s disease (PD), leading to dramatic improvements in quality of life when applied successfully. Unfortunately, outcomes are inconsistent. Suboptimal outcomes likely result from a failure to sufficiently recruit therapeutic neural pathways. Many hypotheses exist as to which neural elements represent the true therapeutic targets for DBS, however a consensus remains elusive. Computational models are promising tools for the theoretical and clinical interpretation of DBS. Patient-specific DBS models have attempted to identify activation ?targets? that correlate with therapeutic response across the population, with the goal of using these targets prospectively to inform the application of DBS. Unfortunately these models have had limited success in predicting patient-specific outcomes. Patient-specific DBS models rely on numerous anatomical, electrical, and biophysical parameters to generate predictions. We used a series of sensitivity analyses to characterize sources of uncertainty in DBS models and evaluate their effects on neural activation predictions. We first quantified the uncertainty associated with postoperative electrode localization, and identified pre-processing steps that have the biggest effect on electrode location. We then identified the effects of electrode location uncertainty on neural activation predictions. We found that even small amounts of electrode location uncertainty (0.5 mm) have dramatic effects on model predictions, and should be minimized through the use of rigorous image registration methods and high-quality, chronic postoperative imaging. Finally, we quantified errors associated with ignoring structural axonal characteristics (specifically termination) by comparing activation thresholds of terminating axons with diameter- matched fibers of passage. Multi-compartment cable models of terminating axons had dramatically lower thresholds than fibers of passage, and ignoring termination underestimates activation of subthalamic afferent fibers. Our results provide potential explanations for some clinical observations, including the superiority of anodic and short pulse width stimulation in providing symptom control. We found that numerous pathways are activated during subthalamic DBS, and identified stimulation approaches which elicit distinct patterns of neural activation. These results inform the next iteration of patient-specific DBS models by providing reasonable estimates of uncertainty in model-predicted neural activation and conceptual approaches to incorporate uncertainty into DBS models.