Motor imagery EEG classification method using 3D CNN and LSTM for rehabilitation application

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Current Brain-Computer Interface (BCI) applications using electroencephalography (EEG) face challenges due to limitations in classification accuracy and robustness.
• Accurate and robust EEG classification is crucial for advancing motor imagery (MI) based BCIs.

Purpose of the Study:

• To develop and evaluate a novel EEG classification method, 3D CNN and LSTM for Motor Imagery (3D-CLMI), to enhance accuracy and robustness in MI tasks.
• To validate the proposed method's performance against state-of-the-art techniques and assess its potential in a practical rehabilitation setting.

Main Methods:

• Proposed 3D-CLMI method combining 3D Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) with attention mechanisms.
• Integrated MI-EEG signals from multiple channels into 3D features, extracting spatial features using multi-scale 3D convolutional kernels.
• Employed a parallel structure to extract spatial and temporal features independently, then combined them for classification.

Main Results:

• Achieved 92.7% classification accuracy and 0.91 F1-score on the BCI Competition IV 2a dataset, outperforming existing methods.
• Demonstrated superior classification accuracy and F1-score on a newly collected dataset from 12 participants performing a four-class MI task.
• Ablation experiments confirmed the effectiveness of individual components within the 3D-CLMI method.

Conclusions:

• The 3D-CLMI method offers significant improvements in accuracy and robustness for MI-EEG signal classification.
• The proposed method shows promise for practical BCI applications, including a VR-based rehabilitation system for patients with impaired hand motor function.