Multibranch convolutional neural network with contrastive representation learning for decoding same limb motor...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) show promise for motor imagery (MI) decoding.
• Classifying MI tasks within the same limb is challenging due to overlapping brain activity.
• Single deep learning models struggle to differentiate similar MI tasks.

Purpose of the Study:

• To enhance the decoding of multiple hand-MI tasks from the same limb.
• To address the limitations of single deep learning models in differentiating similar MI tasks.
• To develop a novel framework for improved MI classification accuracy.

Main Methods:

• A multi-branch convolutional neural network (CNN) framework was proposed.
• The CNN incorporated feature extractors from established deep learning models.
• Contrastive representation learning was employed to derive meaningful feature representations.

Main Results:

• The proposed method achieved 62.98% accuracy with six MI classes and 76.15% with four MI classes.
• Performance surpassed several state-of-the-art methods on benchmark datasets.
• The framework demonstrated superior classification accuracy for challenging hand-MI tasks.

Conclusions:

• The multi-branch CNN framework effectively enhances the decoding of hand-MI tasks.
• The approach shows potential for online BCI applications despite longer training times.
• Careful consideration of trade-offs between model complexity, training, and performance is necessary.