Project 3 will flip around the idea of closed-loop DBS, which has to date been based minimizing or maximizing a particular neurophysiological biomarker. In our case, we will develop unsupervised learning algorithms that can efficiently sample the very broad DBS parameter space using quantitative behavioral metrics of parkinsonian motor signs as the feedback signal. The approach will be developed and evaluated for both STN-DBS and GP-DBS in MPTP-treated parkinsonian models, but has direct translational potential to human PD patients with DBS implants.
The aims include:
1) characterizing the wash-in and wash-out times across the DBS parameter space for each parkinsonian motor sign and then optimizing stimulation patterns to maximize the wash-out times
2) characterizing the therapeutic window (i.e. parameter distance between effective therapy and emergence of motor side-effects) across the DBS parameter space and developing a reinforcement learning algorithm to automatically identify subject-specific parameters that maximize the level of therapy and the therapeutic window
3) perform electrophysiology in the basal ganglia, motor cortex, and brainstem while quantifying the parkinsonian motor signs in the parkinsonian models to identify those electrophysiological features that are predictive of optimized therapy level for each parkinsonian motor sign on a subject-specific basis.