Force decoding using local field potentials in primary motor cortex: PLS or Kalman filter regression?

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) are crucial for controlling external devices.
• Decoding complex movement parameters from neural signals is essential for advanced BCI applications.
• Local Field Potential (LFP) signals from the primary motor cortex (M1) contain movement-related information.

Purpose of the Study:

• To compare the effectiveness of Partial Least Square (PLS) regression and Kalman filters in predicting force parameters from rat M1 LFP signals.
• To evaluate the performance and computational efficiency of these decoding methods.

Main Methods:

• Recorded 16-channel LFP signals from the M1 of rats performing a force-generating behavioral task.
• Applied PLS regression and Kalman filter algorithms to decode the force parameter from the recorded LFPs.
• Assessed decoding performance using correlation coefficient (CC) and normalized mean square error (NMSE).

Main Results:

• Both PLS regression (CC=0.75, NMSE=0.37) and Kalman filters (CC=0.72, NMSE=0.48) effectively decoded force parameters.
• PLS regression outperformed the Kalman filter in both predictive accuracy and computational speed.
• Nonlinear Kalman filters showed similar CC performance to PLS but incurred higher computational costs.

Conclusions:

• PLS regression is a highly effective and efficient linear method for decoding force parameters from M1 LFPs for BCI applications.
• Linear methods like PLS can be superior to more complex nonlinear methods in terms of performance and speed for specific BCI tasks.
• The findings suggest PLS regression is a promising technique for real-world BCI control requiring complex movement decoding.