At-home movement state classification using totally implantable cortical-basal ganglia neural interface

Area of Science:

• Neuroscience
• Biomedical Engineering
• Neuromodulation

Background:

• Decoding human movement from neural signals typically requires lab-based machine learning and short-term data.
• This approach limits understanding of natural behavior and clinical applications like closed-loop neuromodulation.

Purpose of the Study:

• To demonstrate the first in-human, at-home classification of walking using a fully implantable, bidirectional neurostimulator.
• To establish a pipeline for real-world neural decoding and a framework for personalized adaptive neuromodulation.

Main Methods:

• Recorded chronic motor cortex and globus pallidus activity in four Parkinson's disease patients during unsupervised daily activity (over 80 hours).
• Synchronized neural data with wearable kinematic data.
• Identified personalized spectral biomarkers of gait and validated their performance.
• Utilized the neurostimulator's embedded linear discriminant classifier for real-time movement state classification.

Main Results:

• Successfully identified predictive personalized spectral biomarkers for gait.
• Demonstrated real-time movement state classification using these biomarkers and the neurostimulator's classifier.
• Validated the classification performance within device-level constraints for closed-loop stimulation.

Conclusions:

• Established a novel pipeline for real-world neural decoding of human movement.
• Developed a scalable framework for personalized adaptive neuromodulation.
• Expanded the translational potential of implantable brain-computer interfaces for naturalistic behavior.