Stop state classification in intracortical brain-machine-interface.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Invasive brain-machine interfaces (BMIs) offer potential for independent mobility in tetraplegic patients via brain-controlled wheelchairs.
• Robust distinction between stop and movement states is crucial for practical, long-term BMI system use.
• Non-stationarity in neural signals presents a significant challenge for reliable BMI control.

Purpose of the Study:

• To investigate the non-stationarity of the stop state in neural data for BMI control.
• To develop and evaluate a hybrid approach for improved decoding accuracy in brain-controlled movement.

Main Methods:

• Collected neural data from a macaque controlling a robotic platform to stop and move.
• Proposed a hybrid decoding approach combining random forest and linear discriminant analysis (LDA).
• Performed offline decoding over three months, comparing the hybrid method against LDA alone.

Main Results:

• The hybrid approach demonstrated an average performance increase of 22.7% in decoding accuracy.
• Significant improvements were observed during sessions where LDA alone exhibited bias towards the stop state.
• The hybrid method effectively addressed non-stationarity in neural signals, particularly in the stop state.

Conclusions:

• The proposed hybrid approach enhances the accuracy of brain-machine interface control by mitigating non-stationarity.
• This method facilitates more reliable decoding of movement states, crucial for assistive technologies like brain-controlled wheelchairs.
• The findings suggest a pathway towards more robust and practical BMI systems for individuals with severe motor impairments.