Generalizability of EMG decoding using local field potentials.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Robotics

Background:

• Motor cortical local field potentials (LFPs) are used for decoding arm movement kinematics and kinetics.
• Clinical viability of prostheses requires brain activity decoders to generalize across diverse behavioral conditions.
• Generalization capability of LFP-based decoders across different behaviors remains insufficiently demonstrated.

Purpose of the Study:

• To provide evidence for the generalization ability of LFP-based electromyogram (EMG) decoders across two distinct behavioral tasks.
• To assess the performance of LFP-EMG decoders in both within-task and across-task decoding scenarios.
• To evaluate the potential of LFP signals for robust brain-computer interfaces in prosthetic applications.

Main Methods:

• Implanted intracortical microelectrode arrays in primary motor (M1) and ventral pre-motor (PMv) cortices of a rhesus macaque.
• Recorded LFP and EMG activity during a 2-D center-out isometric wrist torque task (TT) and free reach and grasp behavior (FB).
• Trained EMG decoders using temporal and spectral features of LFP signals, with data from both behaviors.

Main Results:

• LFP-based EMG decoders demonstrated generalization across different behavioral tasks.
• The average generalization score achieved was 65 ± 20%.
• Individual decoding scores in some cases reached 100%, indicating strong performance.

Conclusions:

• LFP-based EMG decoders exhibit reasonable generalization capabilities across distinct behaviors.
• This finding represents a significant step towards developing clinically applicable prosthetic devices.
• Further research can build upon these results to enhance decoder robustness for real-world prosthetic use.