Decoding the cortical responses to mechanical wrist perturbations: A two-step shared structure NARX method.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Understanding cortical responses to external stimuli is crucial for brain function research and brain-computer interfaces.
• Existing nonlinear autoregressive with exogenous input (NARX) models often lack accuracy due to neglecting data relationships and using greedy search for regression terms.

Purpose of the Study:

• To develop an improved two-step method for constructing shared structure NARX models of cortical responses.
• To enhance the accuracy and descriptive power of models for brain information encoding and neural dynamics.

Main Methods:

• A two-step approach involving data concatenation from multiple trials.
• Orthogonal Forward Regression (OFR) to identify shared model terms across subjects, forming a general target model.
• Adaptive multi-population genetic algorithms to identify non-shared terms for enhancing model descriptive power.

Main Results:

• The proposed method demonstrates significant accuracy compared to state-of-the-art techniques in simulations.
• Application to electroencephalography (EEG) signals revealed consistent neural response patterns across subjects under mechanical disturbance.
• Analysis suggested subject-specific responses may originate from muscle feedback and the brain generates motor inhibition signals.

Conclusions:

• The novel method effectively models cortical responses and enhances brain information encoding.
• Findings offer insights into neural response mechanisms, including sensory-motor integration and maintaining a resting state.
• The study has potential implications for advancing brain-computer interface technologies and understanding neural dynamics.