STIT-Net- A Wavelet based Convolutional Transformer Model for Motor Imagery EEG Signal Classification in the...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor Imagery (MI) electroencephalographic (EEG) signal classification is crucial for developing advanced mobility rehabilitation technologies.
• Current methods face challenges in accurately capturing complex spatio-temporal dynamics within EEG signals.

Purpose of the Study:

• To propose and evaluate an end-to-end hybrid deep network, the Spatio Temporal Inception Transformer Network (STIT-Net), for enhanced MI classification.
• To leverage Discrete Wavelet Transform (DWT) for extracting dominant alpha and beta frequency bands to improve classification performance.

Main Methods:

• The STIT-Net model integrates spatial and temporal convolutions, an inception block for multi-level feature extraction, and a transformer encoder with a self-attention mechanism.
• Discrete Wavelet Transform (DWT) was applied to isolate alpha (8-13 Hz) and beta (13-30 Hz) EEG sub-bands.

Main Results:

• The STIT-Net model achieved high classification accuracies on the Physionet EEG motor imagery dataset: up to 95.70% for binary classification and 87.34% for three-class classification in the beta band.
• Performance demonstrated superiority over existing literature results, with consistent improvements across alpha and beta bands for varying class numbers.
• The model's robustness was confirmed through evaluations on diverse motor imagery datasets under subject-independent and cross-subject conditions.

Conclusions:

• The STIT-Net model represents a significant advancement in EEG-based MI classification, offering improved accuracy and robustness.
• This deep learning approach holds substantial promise for enhancing the efficacy of brain-computer interfaces in mobility rehabilitation.
• The hybrid architecture effectively captures intricate spatio-temporal features essential for precise motor imagery decoding.