Monte Carlo Dropout for Uncertainty Estimation and Motor Imagery Classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI)-based Brain-Computer Interfaces (BCIs) offer communication for motor-impaired individuals.
• Deep learning advances MI-based BCIs but lacks uncertainty quantification, leading to overconfident predictions.

Purpose of the Study:

• To enhance the reliability of MI-based BCIs for practical applications.
• To introduce methods for quantifying predictive uncertainty in deep learning models for MI tasks.

Main Methods:

• Monte Carlo dropout (MCD) was applied to Shallow Convolutional Neural Network (SCNN) and ensemble models (E-SCNN) for MI classification and uncertainty estimation.
• A threshold approach was developed to identify and manage high-uncertainty predictions from SCNN-MCD and E-SCNN-MCD.

Main Results:

• The proposed SCNN-MCD and E-SCNN-MCD methods improved MI classification accuracy.
• Uncertainty estimation was successfully integrated, providing a more reliable BCI system.
• The threshold approach effectively discriminated high-uncertainty predictions.

Conclusions:

• The integration of Monte Carlo dropout enhances the performance and reliability of deep learning-based MI-BCIs.
• The proposed methods offer a pathway towards more robust and trustworthy BCIs for patients with severe motor disabilities.