Finding the best deep brain stimulation (DBS) settings from many potential combinations takes time and requires highly experienced medical experts. Therefore, researchers created an automated method to identify optimal stimulation settings in patients with Parkinson’s disease (PD) receiving subthalamic nucleus (STN) DBS based on imaging-derived criteria.

A prediction model for beneficial and unfavorable stimulation outcomes was trained using electrode locations and monopolar review data from 612 stimulation settings obtained from 31 Parkinson’s disease patients. The model’s performance was assessed inside the training cohort and on an independent cohort of 19 patients using cross-validation. They inverted the model by using brute force to find the best stimulation locations in the target area. Finally, an optimization method was developed to determine the best stimulation settings. Suggested stimulation settings were compared to clinical practice parameters.

The predicted motor outcome across the training cohort corresponded with the observed outcome (R=0.57, P<10−10). The model explained 28% of the variation in motor outcome differences between settings in the test cohort. The dorsolateral border of the STN was chosen as the stimulation location for maximal motor improvement. Model-based suggestions more closely matched the environment with higher motor progress than the two empirical settings. They created and tested a data-driven algorithm that recommended stimulation levels that result in optimal motor improvement while limiting the likelihood of stimulation-induced adverse effects. The technique may give future guidelines for DBS programming.