Test whether symptom-specific connectomic targets for gait dysfunction can be translated into actionable DBS settings using an optimization algorithm.

Symptom-specific deep brain stimulation (DBS) targets (functional networks or fiber tracts) derived from connectomics are compelling, but it is unclear how to translate them into DBS parameters. Here, we develop a machine learning algorithm which translates the symptom-specific targets into DBS parameters. We investigate this using gait dysfunction in Parkinson's and evaluate two recent gait-specific targets. 

We built an optimization algorithm that individualizes each electrode’s parameters to maximize overlap of the DBS stimulation with symptom-specific targets. We used two retrospective cohorts (training n=44; test n=100) as well as a recent gait-specific brain network and gait-specific fiber tract. We developed a hand-crafted optimization algorithm and tuned it on the training cohort, then tested it using the held-out test cohort. We assessed if: 1) the algorithm provided significantly different DBS parameters than best clinical settings, 2) if being closer to theoretically optimal DBS parameters was associated with better gait scores, and 3) if the algorithm provided useful settings in a small prospective pilot of 4 patients.

Optimizer-suggested programs showed markedly greater engagement of both targets compared to the best clinical settings (functional: t=27.3, p<0.0001; tract: t=17.0, p<0.0001). Further, increased similarity to the 'gait-optimal' settings was associated with gait improvement one year after DBS for both the functional target (p=0.37, p<0.0001) and the fiber tract target (p=0.52, p<0.0001). Finally, in a prospective feasibility step (n=4), reprogramming to the optimizer-derived settings improved gait without adverse effects.