Machine Learning Will Extend the Clinical Utility of Adaptive Deep Brain Stimulation.

Member Authors
Neumann WJ, Rodriguez-Oroz MC.


Sensing-enabled implantable pulse generators for deep brain stimulation (DBS) now make it possible to integrate invasive brain signal recordings into clinical routine. Decades of research have characterized the intricate relationship between pathological neural circuit activity and movement disorder phenomenology. Motor state-specific pathological activity has been identified in various DBS implantation targets in patients who suffer from Parkinson’s disease (PD), dystonia, Tourette syndrome, and essential tremor (ET). The fact that neural oscillations can convey information on concurrent motor signs has inspired the idea of closed-loop adaptive deep brain stimulation (aDBS). Closed-loop aDBS is designed to adapt stimulation dynamically to feedback signals and is thus contrasted with conventional open-loop DBS, where chronic stimulation parameters need be changed manually by specialized neurologists or patients and are kept constant throughout changes in medication efficacy as observed in PD patients with dopaminergic drugs. For aDBS, the presence or threshold crossing of a specific activity pattern is envisioned as the feedback signal to adapt stimulation parameters dynamically. For tremor, earlier studies have focused on kinematic measures using external accelerometers to adapt stimulation amplitude or phase. Such a single biomarker approach has the advantage that the feedback signal simultaneously represents a pathophysiologically relevant target feature. However, kinematic markers require motor signs to be present, and thus they may signal therapeutic demand later than necessary. Beyond tremor, most identified oscillatory biomarkers have been associated with specific motor signs within broader syndromes of complex neural system disorders that would benefit from complementary adaptation measures. These caveats can possibly be mitigated with machine learning toward a precision medicine approach for intelligent real-time aDBS.

Published: Apr 2021