Logistic-weighted regression improves decoding of finger flexion from electrocorticographic signals

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCIs) are advancing, particularly in decoding limb movements using invasive techniques like electrocorticography (ECoG).
• Current decoding methods often use regression models for continuous movement trajectories, potentially overlooking the inherent binary states of motion and rest.

Purpose of the Study:

• To propose and evaluate a novel algorithm, logistic-weighted regression, for enhanced movement decoding in BCIs.
• To address the overlooked binary property (motion vs. rest) in limb movement decoding.

Main Methods:

• Development and application of a logistic-weighted regression algorithm.
• Decoding human finger flexion from ECoG signals using the proposed algorithm.
• Comparison of logistic-weighted regression against linear and pace regression models.

Main Results:

• Logistic-weighted regression significantly improved decoding performance compared to linear and pace regression.
• The algorithm effectively utilizes the binary motion/rest state property in movement decoding.
• Enhanced decoding accuracy for finger flexion from ECoG signals.

Conclusions:

• Logistic-weighted regression offers superior performance for BCI movement decoding by incorporating state information.
• The proposed algorithm is a valuable tool for BCIs decoding continuous movements, including finger flexion.
• This approach has broad applicability in advancing BCI technology for movement prediction.