Long-term unsupervised recalibration of cursor-based intracortical brain-computer interfaces using a hidden Markov...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Intracortical brain-computer interfaces (iBCIs) require frequent recalibration due to neural activity drift.
• This drift leads to performance degradation and usability issues for iBCI users.
• Current recalibration methods face challenges in maintaining long-term performance.

Purpose of the Study:

• To develop an unsupervised adaptation method for iBCIs.
• To improve the robustness and longevity of iBCI performance.
• To overcome a major barrier in the clinical translation of BCIs.

Main Methods:

• Introduction of a hidden Markov model (HMM) for inferring user targets during iBCI use.
• Retraining the iBCI system using inferred targets for unsupervised adaptation.
• Comparison with distribution alignment methods in simulations and human user studies.

Main Results:

• The HMM-based approach outperformed distribution alignment methods in long-term closed-loop simulations and human use.
• The method demonstrated capability for long-term unsupervised recalibration on a five-year iBCI dataset.
• In contrast, data-distribution-matching approaches showed accumulating errors over time.

Conclusions:

• The proposed target inference recalibration method enables robust, long-term unsupervised adaptation in iBCIs.
• Leveraging task structure significantly enhances decoder performance.
• This approach addresses a critical limitation for the clinical translation of brain-computer interfaces.