Robust Local Field Potential-based Neural Decoding by Actively Selecting Discriminative Channels.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Local field potentials (LFPs) are explored as neural decoding signals for brain-machine interfaces (BMIs).
• Redundancy in LFP signals across frequency bands and channels can impede decoding performance.
• Spike signal deterioration in invasive BMIs necessitates alternative neural signals.

Purpose of the Study:

• To develop and evaluate a novel method for selecting discriminative LFP channels.
• To reduce redundancy in LFP signals for improved neural decoding.
• To maximize class separation using a Fisher-distance ratio-based channel selection technique.

Main Methods:

• Proposed a Fisher-distance ratio-based method for active batch selection of discriminative LFP channels.
• Utilized data from a non-human primate recorded over 5 non-consecutive days.
• Trained decoding models using initial sessions and tested on subsequent sessions for 4-class decoding.

Main Results:

• Achieved average decoding accuracies of 79.55% (beta), 79.02% (low gamma), and 79.40% (high gamma) using selected LFP channels.
• Selected LFP channels in the high gamma band increased decoding accuracy by 8.67% compared to using all LFP channels.
• Decoding accuracy using selected LFP channels was 7.26% lower than spike-based decoding.

Conclusions:

• The proposed Fisher-distance ratio method effectively selects discriminative LFP channels for neural decoding.
• Channel selection significantly enhances decoding performance by mitigating LFP signal redundancy.
• This method offers a promising approach to improve BMI performance using LFP signals.