MACNet: A Multidimensional Attention-Based Convolutional Neural Network for Lower-Limb Motor Imagery Classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Decoding lower-limb motor imagery (MI) is vital for brain-computer interfaces (BCIs) and rehabilitation engineering.
• Classifying lower-limb MI from electroencephalogram (EEG) signals is difficult due to low-quality signals and physiological similarities.

Purpose of the Study:

• To propose a novel multidimensional attention-based convolutional neural network (CNN), named MACNet, for enhanced lower-limb MI classification.
• To address the challenges of low signal quality and complex feature extraction in lower-limb MI decoding.

Main Methods:

• Developed MACNet, integrating temporal refining and attention-enhanced convolutional modules to leverage CNNs and attention mechanisms.
• Utilized a newly created lower-limb MI dataset and the BCI Competition IV 2a dataset for comprehensive evaluation.
• Conducted comparison experiments and ablation studies to validate model performance.

Main Results:

• MACNet achieved state-of-the-art performance in subject-specific lower-limb MI classification.
• Outperformed existing models on both the custom and public EEG datasets.
• Visualization analysis demonstrated MACNet's robust feature learning and potential insights into brain activity related to lower-limb MI.

Conclusions:

• MACNet effectively decodes lower-limb motor imagery from EEG signals, offering a significant advancement for BCIs and rehabilitation.
• The model's design enhances feature extraction, leading to superior classification accuracy.
• MACNet demonstrates strong generalizability and effectiveness, verified through rigorous experimental validation.