A Deep Convolutional Neural Network Based Classification Of Multi-Class Motor Imagery With Improved Generalization.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor imagery (MI) based brain-computer interfaces (BCI) are vital for rehabilitation and prosthetics.
• Current MI-BCI systems struggle with interpersonal variability, especially in multi-class scenarios.
• Handcrafted features limit the generalization capabilities of existing MI interpretation models.

Purpose of the Study:

• To develop an end-to-end Convolutional Neural Network (CNN) model for improved motor imagery classification.
• To address interpersonal variability in MI-BCI by employing a novel feature extraction and filtering approach.
• To enhance the generalizability and real-time classification performance of MI-BCI systems.

Main Methods:

• Utilized a CNN-based model incorporating axis shuffling for 1D preprocessing and parameter reduction.
• Implemented end-to-end training for integrated filtering and feature extraction.
• Evaluated the model on the publicly available BCI Competition-IV 2a dataset.

Main Results:

• The proposed CNN model achieved an average accuracy of 70.5% and a highest accuracy of S3.6%.
• Demonstrated the capability to identify subject-specific frequency bands for enhanced MI classification.
• The model performs real-time classification without requiring GPU acceleration.

Conclusions:

• The developed CNN model effectively mitigates interpersonal variability in MI-BCI.
• The axis shuffling technique improves model generalization and reduces overfitting.
• This approach offers a promising solution for real-time, accurate MI classification in practical applications.