CTSSP: A temporal-spectral-spatial joint optimization algorithm for motor imagery EEG decoding

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor imagery brain-computer interfaces (MI-BCIs) show promise for neurorehabilitation.
• EEG signal challenges like non-stationarity, low SNR, and cross-session variability limit current MI-BCI performance.
• Existing decoding methods often isolate temporal, spectral, and spatial feature optimization, leading to suboptimal results.

Purpose of the Study:

• To develop a unified framework for jointly optimizing temporal, spectral, and spatial features in MI-BCIs.
• To enhance the robustness and accuracy of EEG decoding in challenging neurorehabilitation scenarios.
• To address the limitations of fragmented optimization in current MI-BCI decoding methods.

Main Methods:

• Proposed Common Temporal-Spectral-Spatial Patterns (CTSSP), a unified framework for joint filter optimization.
• Integrated multi-scale temporal segmentation for dynamic neural evolution capture.
• Employed channel-adaptive FIR filters and low-rank regularization for enhanced rhythm detection and generalization.

Main Results:

• CTSSP achieved state-of-the-art performance across five public datasets.
• Demonstrated superior accuracies: 76.9% (within-subject), 68.8% (cross-session), and 69.8% (cross-subject).
• Significantly outperformed baselines and proved competitive against deep learning models, with learned filters aligning with motor cortex mechanisms.

Conclusions:

• CTSSP effectively extracts robust, interpretable, and coupled spatio-temporal-spectral patterns.
• Provides a powerful, data-efficient solution for decoding MI EEG in noisy, non-stationary conditions.
• The developed framework overcomes limitations of decoupled feature extraction for improved neurorehabilitation outcomes.