Personalized computational models of deep brain stimulation

Abstract

Deep brain stimulation (DBS) therapy is used for managing symptoms associated with a growing number of neurological disorders. One of the primary challenges with delivering this therapy, however, continues to be accurate neurosurgical targeting of the DBS lead electrodes and post-operative programming of the stimulation settings. Two approaches for addressing targeting have been advanced in recent years. These include novel DBS lead designs with more electrodes and computational models that can predict cellular modulation during DBS. Here, we developed a personalized computational modeling framework to (1) thoroughly investigate the electrode design parameter space for current and future DBS array designs, (2) generate and evaluate machine learning feature sets for semi-automated programming of DBS arrays, (3) study the influence of model parameters in predicting behavioral and electrophysiological outcomes of DBS in a preclinical animal model of Parkinson’s disease, and (4) evaluate feasibility of a novel endovascular targeting approach to delivering DBS therapy in humans. These studies show how independent current controlled stimulation with advanced machine learning algorithms can negate the need for highly dense electrode arrays to shift, steer, and sculpt regions of modulation within the brain. Additionally, these studies show that while advanced and personalized computational models of DBS can predict many of the behavioral and electrophysiological outcomes of DBS, there are remaining inconsistencies that suggest there are additional physiological mechanisms of DBS that are not yet well understood. Finally, the results show how computational models can be beneficial for prospective development of novel approaches to neuromodulation prior to large-scale preclinical and clinical studies.