Penalized Multi-Way Partial Least Squares for Smooth Trajectory Decoding from Electrocorticographic (ECoG) Recording.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor-related Brain-Computer Interface (BCI) systems are crucial for restoring function.
• Accurate and smooth prediction of upper limb trajectories is a key challenge.
• Existing methods like Kalman Filters and standard PLS have limitations in prediction accuracy, smoothness, or delay.

Purpose of the Study:

• To propose novel decoding algorithms for continuous upper limb trajectory prediction in motor-related BCI systems.
• To compare the performance of proposed methods against established techniques.
• To evaluate prediction accuracy, trajectory smoothness, and prediction delay.

Main Methods:

• Developed two new prediction methods: Sobolev Penalized Multi-Way Partial Least Squares (PLS) and Polynomial Penalized Multi-Way Partial Least Squares (PLS) regressions.
• Compared these methods against standard Multi-Way Partial Least Squares (PLS) and Kalman Filter approaches.
• Evaluated algorithms based on prediction accuracy, trajectory smoothness, and prediction delay.

Main Results:

• The proposed Sobolev and Polynomial Penalized Multi-Way PLS methods achieved prediction accuracy comparable to other PLS algorithms.
• These novel methods significantly improved trajectory smoothness, rivaling the performance of the Kalman Filter.
• A key finding is the substantially reduced prediction delay compared to the Kalman Filter approach.

Conclusions:

• The proposed Sobolev and Polynomial Penalized Multi-Way PLS algorithms offer a superior balance of prediction accuracy, trajectory smoothness, and reduced delay for motor-BCI systems.
• These advanced decoding algorithms have potential applications beyond neuroscience, including robotics and human-computer interaction.
• The findings represent a significant advancement in BCI technology for upper limb movement prediction.