EEG-SGENet: A lightweight convolutional network integrating SGE for motor imagery brain-computer interfaces

Area of Science:

• Neuroscience
• Computer Science
• Machine Learning

Background:

• Deep learning models for electroencephalography (EEG)-based motor imagery brain-computer interfaces (MI-BCI) achieve high accuracy but suffer from increasing model size and computational demands.
• Balancing model accuracy with computational efficiency is a persistent challenge in MI classification research.
• Convolutional Neural Networks (CNNs) are susceptible to noise, impacting feature representation.

Purpose of the Study:

• To introduce a novel, lightweight end-to-end CNN model, EEG-SGENet, that enhances both accuracy and computational efficiency for MI-BCI.
• To address the limitations of existing models by incorporating a Spatial Group-wise Enhance (SGE) module to improve feature extraction and reduce noise.
• To achieve a balance between decoding performance and computational cost in EEG signal analysis.

Main Methods:

• Proposed EEG-SGENet, a novel end-to-end convolutional neural network incorporating the lightweight Spatial Group-wise Enhance (SGE) module.
• The SGE module enhances useful features and suppresses noise by generating attention factors for spatial locations within semantic groups.
• Evaluated the model on the BCI IV 2a and BCI IV 2b datasets for motor imagery classification tasks.

Main Results:

• EEG-SGENet achieved 80.98% accuracy for four-category motor imagery classification on the BCI IV 2a dataset.
• The model obtained an average classification accuracy of 76.17% for two-category tasks on the BCI IV 2b dataset.
• Comparisons demonstrated that EEG-SGENet offers a superior balance between decoding performance and computational cost compared to other lightweight models.

Conclusions:

• EEG-SGENet represents a significant advancement in developing efficient and accurate MI-BCI systems.
• The proposed model effectively balances high classification accuracy with reduced computational requirements.
• EEG-SGENet is a promising new method for decoding electroencephalography signals in brain-computer interfaces.