2147 Stable Intracortical Brain-Computer Interfaces in People With Motor Disabilities

View at publisher Summarise with VIRAT
Authors: Matthew J. Mender, Joseph T. Costello, Matthew S. Willsey
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

INTRODUCTION: Motor intracortical brain-computer interfaces (iBCIs) are promising for restoring motor function to people with paralysis. These systems decode motor intent from neural activity which can then be used to control assistive devices. The decoding algorithms require calibration each session to align neural activity with intended movements. Studies with reaching and grasping have shown that this recalibration burden can be reduced if a stable neural representation of movements is modeled and used for decoding. Extending these results to multiple individuated fingers could allow fine motor restoration. METHODS: Finger movements and intracortical recordings from the precentral gyrus were collected from a nonhuman primate across 12 sessions spanning 427 days. A multiday latent factor analysis via dynamical systems (LFADS) model, incorporating session-specific read-in and read-out layers, was trained to identify stable latent factors. An additional LFADS model with a static read-in network using channel statistics (mean and covariance) was also trained to identify latent factors with no session-specific recalibration. We evaluated stability by decoding finger movements from smoothed spiking activity or LFADS-calculated latent factors. RESULTS: Smoothed-spike decoders exhibited significant performance decay when evaluated on future sessions (correlation drop: 5.93e-4/day; p = 1.2e-8). In contrast, LFADS-calculated factors allowed consistent decoding in future sessions (correlation drop: -9.8e-5/day; p = 0.07). Using a static read-in network improved prediction accuracy in sessions held-out from LFADS training (delta corr = +0.1; p = 7e-3). CONCLUSIONS: These findings demonstrate the potential to find stable, cross-day latent representations for finger movements that could reduce the need for daily recalibration in iBCIs.

Fields of Study & Subjects

EEG and Brain-Computer Interfaces Neuroscience and Neural Engineering Motor Control and Adaptation

Publication details

Journal / source
Neurosurgery
Volume
72
Issue
Supplement_1
Pages
286–286
Publication date
26 March 2026
Language
EN