Seizure characterisation using frequency-dependent multivariate dynamics

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Epileptic seizure characterization is crucial for developing effective treatments and surgical planning.
• Multivariate techniques are increasingly used to analyze complex electroencephalographic (EEG) data.
• Understanding cross-channel dynamics in EEG can provide insights into brain activity during seizures.

Purpose of the Study:

• To investigate frequency-dependent cross-correlation dynamics between EEG channels during epileptic seizures.
• To explore the utility of eigenspectrum analysis of cross-correlation matrices for seizure detection.
• To identify dynamic changes in EEG signal characteristics associated with seizure activity.

Main Methods:

• Applied the Maximum Overlap Discrete Wavelet Transform (MODWT) to decompose EEG signals into different frequency bands.
• Analyzed the dynamics of the cross-correlation matrix between EEG channels at each frequency using eigenspectrum analysis.
• Examined the distribution of wavelet energy across frequencies during seizure events.

Main Results:

• Identified frequency-dependent changes in EEG channel correlation structure, indicative of seizure activity.
• Observed increased correlations between channels at higher frequencies during seizures.
• Detected a redistribution of wavelet energy, with higher fractional energy in high frequencies during seizures.
• Noted dynamical changes in both correlation and energy at lower frequencies during seizures.

Conclusions:

• The proposed method, analyzing frequency-dependent correlation structure and energy distribution, can characterize changes in EEG signals during epileptic seizures.
• Eigenspectrum analysis of MODWT-derived cross-correlation matrices offers a promising approach for seizure characterization.
• Further research into eigenvalues and inter-frequency correlations may reveal additional seizure characteristics.