Neural Noise- Adaptive DBS for Parkinson’s disease can be driven by a remotely optimized neural decoder that is movement-responsive.1
- The approach ties neural decoding to closed-loop neuromodulation and supports near-term clinical translation.1 1
Gardner updates
- Movement-responsive DBS for Parkinson’s using a remotely optimized neural decoder was reported in Nature, tying decoding to closed-loop neuromodulation and near-term clinical translation. 1
Weekly enrichment (2026-07-20)
- The study reports the first fully embedded movement-responsive adaptive DBS (aDBS) in a single human with Parkinson’s disease, implanted bilaterally with the Medtronic Summit RC+S system; each hemisphere combined an STN depth lead with a subdural ECoG strip over the precentral and postcentral gyri.2 3
- The onboard controller used linear discriminant analysis over up to four LFP power bands per hemisphere, and feature-importance analysis showed cortical (ECoG) signals were the most valuable inputs for decoding movement.2
- Personalized, PCA/STFT-derived power bands outperformed canonical frequency bands by roughly 6% (left) and 2-3% (right) in classification accuracy (p<0.001, permutation test).2
- On a held-out session, offline movement-decoding accuracy was 83% (left hemisphere) and 76% (right), with true-positive rates of 88%/93% and true-negative rates of 81%/68% (all p<0.001).2
- Decoders were remotely optimized in the participant’s home using Gaussian-process Bayesian optimization on ~25-minute daily sessions with Apple Watch accelerometry labels and five-fold (per-day) cross-validation, with a separate day held out for testing.2
- Across 12 blinded sessions spanning 154 days (final session 435 days after training), movement-responsive aDBS significantly reduced resting dyskinesia measured as 1-4 Hz accelerometry power versus both constant and inverted control conditions (p=8.2e-6 vs constant; p=1.1e-6 vs inverted).2
- Movement-locked stimulation shortened keypress duration (ANCOVA p=4.6e-7) and increased typing speed by about 0.33 keypresses/s versus the inverted control, with faster dominant-hand repetition rates and no significant change in error (backspace) rate.2
- Stimulation alternated between 1.6 mA at rest and 2.2 mA during movement around a 1.9 mA constant reference; therapeutic self-scores tracked classifier F1 (Pearson r=0.52) and correlated negatively with mean current (r=-0.44), arguing benefits came from accurate targeting rather than more stimulation.2
- The peer-reviewed version appeared in Nature Biomedical Engineering (June 2025, Dixon et al.) under IDE G180097 and ClinicalTrials.gov registration NCT03582891.3
- Broader context: the pivotal ADAPT-PD trial (JAMA Neurology 2025) and a Nature Medicine gait-phase aDBS feasibility trial (2026, NCT04675398) indicate adaptive DBS is reaching clinical viability, complementing this movement-responsive decoder approach.4 5