Optimization Algorithms for Spatially Targeted Deep Brain Stimulation

Abstract

Optimization algorithms hold significant promise for precision medicine. This dissertation focuses on the application of optimization algorithms to improve the efficacy and efficiency of deep brain stimulation (DBS) therapy for treating brain disorders. Targeting of DBS therapy for a given patient involves neurosurgical implantation of one or more leads of electrodes within the brain and then identifying a set of electrode configurations and stimulation amplitudes that most robustly suppress clinical symptoms. One approach to improve spatial targeting of DBS therapy has been the development of DBS leads with multiple electrodes positioned around and along the lead implant. However, with the additional number of electrodes, brute-force determination of stimulation settings that can most selectively modulate the brain pathways of interest becomes a time-consuming challenge. This is especially relevant since DBS can also induce adverse side effects such as involuntary muscle contractions and mood changes if stimulation is not delivered correctly. In this dissertation, I will show how convex and particle swarm optimization techniques using multi-objective contexts can be applied to subject-specific computational models of DBS to address this challenge. These optimization algorithms leverage complex bioelectric tissue models as well as detailed anatomical and biophysical computational models of the motor thalamus for treating Essential Tremor and the subthalamic nucleus region for treating Parkinson’s disease. Both convex and particle swarm optimization demonstrated robust and efficient performance in generating spatially targeted solutions consisting of non-trivial combinations of active electrodes and stimulation amplitudes. These optimization algorithms have important applications for both targeting specific brain pathways to understand their role in behavior as well as in helping clinicians reduce the dimensionality of the stimulation parameter feature space to evaluate.