Deep Neural Regression Prediction of Motor Imagery Skills Using EEG Functional Connectivity Indicators.

Area of Science:

• Neuroscience
• Rehabilitation Engineering
• Computational Neuroscience

Background:

• Motor imagery (MI) promotes neuroplasticity and recovery in neurophysical regulation.
• Many users struggle with sensorimotor cortex control, impacting MI effectiveness.
• Understanding brain states and functional connectivity is crucial for optimizing MI.

Purpose of the Study:

• To develop an advanced deep network regression model for predicting neurophysiological inefficiency during motor imagery practice.
• To overcome challenges in predicting MI inefficiency, such as overfitting in neural network regression.
• To enhance the prediction performance and physiological interpretability of MI-related applications.

Main Methods:

• Implemented a deep network regression model incorporating leave-one-out cross-validation with Monte Carlo dropout.
• Utilized subject clustering for MI inefficiency and transfer learning between runs.
• Validated the model using functional connectivity predictors from two electroencephalographic (EEG) databases (150 users).

Main Results:

• Achieved high prediction accuracy for pretraining desynchronization and initial training synchronization.
• Demonstrated adequate physiological interpretability of the prediction model.
• Successfully addressed the overfitting risk in neural network regression for MI inefficiency prediction.

Conclusions:

• The developed deep network regression model significantly improves the prediction of motor imagery neurophysiological inefficiency.
• The integration of cross-validation, dropout, clustering, and transfer learning offers a robust approach for MI applications.
• This method holds promise for personalized neurorehabilitation by identifying and addressing user-specific challenges in motor imagery practice.