Objective: Open questions remain regarding the optimal target, or sweetspot, for deep brain stimulation (DBS) in e.g. Parkinson's Disease. Previous studies introduced different methods of mapping DBS effects to determine sweetspots. While having a direct impact on surgical targeting and postoperative programming in DBS, these methods so far have not been investigated in ground truth data.
Materials & Methods: This study investigated five previously published DBS mapping methods regarding their potential to correctly identify a ground truth sweetspot. Methods were investigated in silico in eight different use case scenarios, which incorporated different types of clinical data, noise, and differences in underlying neuroanatomy. Dice coefficients were calculated to determine the overlap between identified sweetspots and the ground truth. Additionally, out of sample predictive capabilities were assessed using the amount of explained variance R-squared.
Results: The five investigated methods resulted in highly variable sweetspots. Methods based on voxel-wise statistics against average outcomes showed the best performance overall. While predictive capabilities were high, even in the best of cases Dice coefficients remained limited to values around 0.5, highlighting the overall limitations of sweetspot identification.
Conclusions: This study highlights the strengths and limitations of current approaches to DBS sweetspot mapping. Those limitations need to be taken into account when considering the clinical implications. All future approaches should be investigated in silico before being applied to clinical data.