Multiclass classification of motor imagery tasks based on multi-branch convolutional neural network and temporal...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor imagery (MI) involves mentally rehearsing movements without physical execution.
• MI-based brain-computer interfaces (BCIs) show promise but face challenges in accurate decoding and understanding neural mechanisms.
• Current limitations hinder the clinical application and development of MI-BCI systems.

Purpose of the Study:

• To develop a novel deep learning framework for decoding multi-class motor imagery (MI) tasks.
• To enhance the accuracy and effectiveness of MI-based brain-computer interface (BCI) systems.
• To address the challenges in decoding MI signals and understanding underlying neural mechanisms.

Main Methods:

• Proposed a multi-branch convolutional neural network (MBCNN) with a temporal convolutional network (TCN) as an end-to-end deep learning framework.
• Utilized MBCNN with diverse convolutional kernels to capture temporal and spectral information from MI electroencephalography (EEG) signals.
• Employed TCN to extract more discriminative features from the MI EEG data.

Main Results:

• Achieved an average accuracy of 75.08% for 4-class motor imagery (MI) task classification on the BCI Competition IV-2a dataset.
• The proposed MBCNN-TCN-Net framework outperformed several state-of-the-art approaches in MI task decoding.
• Demonstrated effective capture of discriminative features for improved MI-BCI performance.

Conclusions:

• The MBCNN-TCN-Net framework effectively decodes motor imagery (MI) tasks by capturing discriminative features.
• The developed approach shows significant potential for improving the performance and clinical applications of MI-based brain-computer interfaces (BCIs).
• Findings contribute to advancing the development of more robust and clinically viable MI-BCI systems.