Bayesian Optimization of Neurostimulation (BOONStim)

Abstract

BackgroundTranscranial magnetic stimulation (TMS) treatment response is influenced by individual variability in brain structure and function. Sophisticated, user-friendly approaches, incorporating both established functional magnetic resonance imaging (fMRI) and TMS simulation tools, to identify TMS targets are needed. ObjectiveThe current study presents the development and validation of the Bayesian Optimization of Neuro-Stimulation (BOONStim) pipeline. MethodsBOONStim uses Bayesian optimization for individualized TMS targeting, automating interoperability between surface-based fMRI analytic tools and TMS electric field modeling. BOONStims Bayesian optimization performance was evaluated in a sample dataset (N=10) using standard circular and functional connectivity-defined targets, and compared to densely sampled grid optimization. ResultsBayesian optimization converged to similar levels of total electric field stimulation across targets in under 30 iterations, converging within 5% error of the maxima detected by grid optimization, and requiring less time. ConclusionsBOONStim is a scalable and configurable user-friendly pipeline for individualized TMS targeting with quick turnaround.