Hybrid decoding of both spikes and low-frequency local field potentials for brain-machine interfaces

Area of Science:

• Neuroscience
• Biomedical Engineering
• Neural Engineering

Background:

• Current brain-machine interfaces (BMIs) rely on intracortical spikes, which can degrade over time, limiting device lifespan.
• Local field potentials (LFPs) offer a more robust neural signal but have historically yielded lower BMI performance compared to spikes.
• Maximizing BMI utility requires exploring and optimizing diverse neural signal sources.

Purpose of the Study:

• To develop and evaluate a biomimetic BMI driven by low-frequency local field potentials (LFPs).
• To assess the performance of an LFP-based BMI in a non-human primate model.
• To investigate a hybrid BMI approach combining LFP and spike signals for improved decoding accuracy.

Main Methods:

• Implemented a biomimetic decoding strategy utilizing low-frequency LFP signals recorded from the motor cortex.
• Trained a rhesus monkey to control a cursor to acquire and hold randomly placed targets using the LFP-driven BMI.
• Developed and tested a hybrid BMI decoder integrating both LFP and spike data for cursor velocity decoding.

Main Results:

• The LFP-driven BMI achieved a 99% success rate in acquiring and holding targets, representing the highest performance for an LFP-based BMI to date.
• While LFP-driven performance was lower than spike-driven BMIs, it demonstrated significant potential as a control signal.
• The hybrid BMI decoder, incorporating both LFP and spikes, outperformed spike-only decoding, indicating synergistic signal integration.

Conclusions:

• Low-frequency LFPs can effectively drive a high-performance BMI, offering a potentially more stable alternative to spike signals.
• Hybrid BMIs that integrate LFP and spike data can achieve superior performance compared to using either signal alone.
• LFPs hold promise for extending the functional lifespan of BMIs, either as a standalone signal or in conjunction with spikes.