A Dual-Branch Spatiotemporal Framework with Dynamic Weighted Permutation Entropy for Short-Window Motor Imagery EEG...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Decoding electroencephalography (EEG) signals is crucial for low-latency brain-computer interfaces (BCIs).
• Existing models face challenges with high cross-subject variability and low signal-to-noise ratios in short EEG windows.
• Robust feature extraction remains a significant hurdle for practical BCI applications.

Purpose of the Study:

• To develop a spatiotemporal decoding framework for improved EEG signal analysis in BCIs.
• To address limitations in current models regarding cross-subject variability and signal quality.
• To enhance the accuracy and efficiency of short-window EEG decoding for near-online BCI applications.

Main Methods:

• Proposed a novel framework integrating dynamic weighted permutation entropy (DWPE) with a hybrid neural network.
• Introduced DWPE to quantify nonlinear dynamic complexity while preserving amplitude information.
• Employed a cascaded convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) architecture with spatial attention for feature extraction.

Main Results:

• Achieved an average accuracy of 84.35% and an AUC of 0.8821 on the hBCI dataset, significantly outperforming baseline methods (p < 0.01).
• Demonstrated a 3.89% accuracy improvement by integrating DWPE with the spatiotemporal backbone.
• Reported a single-sample inference time of 20.94 ms and a total decision latency of approximately 3.02 s.

Conclusions:

• The proposed spatiotemporal decoding framework offers a favorable balance between decoding accuracy and computational efficiency.
• This method is well-suited for short-window and near-online brain-computer interface applications.
• The integration of DWPE and a hybrid neural network effectively addresses challenges in EEG signal decoding.