Denoising of Ictal EEG Data Using Semi-Blind Source Separation Methods Based on Time-Frequency Priors

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Muscle activity in electroencephalogram (EEG) data mimics low-amplitude ictal discharges, complicating epilepsy diagnosis.
• Time-frequency domain analysis reveals distinct characteristics of ictal signals, unlike time-domain similarities with muscle artifacts.

Purpose of the Study:

• To develop and evaluate novel semi-blind source separation methods for denoising ictal EEG signals.
• To leverage time-frequency signatures of ictal discharges as a priori information for improved source separation.

Main Methods:

• Canonical Correlation Analysis (CCA) was used to extract time-frequency signatures of ictal sources.
• Two novel time-frequency based semi-blind source separation methods, Time-Frequency-Generalized EigenValue Decomposition (TF-GEVD) and Time-Frequency-Denoising Source Separation (TF-DSS), were proposed.
• Performance was evaluated against CCA and Independent Component Analysis (ICA) using simulated and real ictal EEG data.

Main Results:

• The proposed TF-GEVD and TF-DSS methods demonstrated superior performance in denoising ictal EEG signals compared to CCA and ICA.
• Time-frequency signatures effectively guided the separation of ictal signals from muscle artifacts.

Conclusions:

• Time-frequency based semi-blind source separation offers a significant advancement in denoising ictal EEG data.
• The proposed TF-GEVD and TF-DSS methods provide a more effective approach for analyzing epileptic disorders by improving signal clarity.