Performance of blind source separation in localizing seizure onset from routine scalp-EEG

Area of Science:

• Neuroscience
• Medical Imaging
• Signal Processing

Background:

• Ictal source imaging is crucial for localizing the seizure onset zone (SOZ).
• Blind source separation (BSS) techniques for decomposing mixed signals remain underexplored in this context.
• Routine low-density scalp EEG is widely available but has limitations in precise source localization.

Purpose of the Study:

• To evaluate the efficacy of BSS in extracting ictal components for SOZ localization.
• To assess BSS performance using routine low-density scalp EEG data across different settings.
• To compare various BSS algorithms and time window durations for optimal SOZ localization.

Main Methods:

• Analyzed 20 seizures from 10 patients with stereo-EEG-defined SOZs.
• Applied FastICA, InfoMax, and SOBI algorithms over 10-, 30-, and 300-second pre-ictal periods.
• Categorized components as ictal or non-ictal and identified the best-of-BSS and unique ictal components.
• Estimated localization distances and assessed sublobar concordance with the SOZ.

Main Results:

• Sublobar SOZ localization was achieved in 56.7% of unique ictal components (28.9 mm mean distance).
• Best-of-BSS components localized the SOZ in 13/20 seizures (median distance 29.9 mm).
• The InfoMax algorithm using 300-second pre-ictal periods yielded the highest performance (F-scores).

Conclusions:

• BSS, particularly the InfoMax algorithm with extended time windows, can effectively localize the seizure onset zone (SOZ) using routine scalp EEG.
• This approach offers a promising, non-invasive method for improving SOZ identification in epilepsy patients.
• Further refinement of BSS parameters can enhance diagnostic accuracy in ictal source imaging.