Automatic selection of epileptic independent fMRI components

Area of Science:

• Neuroimaging
• Epilepsy Research
• Computational Neuroscience

Background:

• Electroencephalography (EEG)-correlated functional Magnetic Resonance Imaging (fMRI) aids in localizing epileptic activity.
• Limitations in EEG reliability necessitate purely fMRI-based data-driven techniques.
• Independent Component Analysis (ICA) has shown potential in identifying epileptic networks from fMRI, even in EEG-negative cases.

Purpose of the Study:

• To develop an automated technique for selecting epileptic sources from fMRI data.
• To enable prospective clinical application of data-driven fMRI analysis for epilepsy.
• To identify seizure onset zones (SOZ) using fMRI-derived components.

Main Methods:

• Utilized independent component analysis (ICA) to extract sources from fMRI data.
• Implemented a two-step classifier cascade: artifact source removal followed by epileptic source selection.
• Employed four discriminative features to characterize and select epileptic sources among other BOLD-related components.

Main Results:

• The automated technique achieved 77% specificity in identifying epileptic sources.
• Concordance with EEG-correlated fMRI activation maps or SOZ was observed in 71% of cases.
• Demonstrated the utility of ICA in EEG-negative epilepsy cases for source localization.

Conclusions:

• The developed automated technique is valuable for identifying epileptic sources using fMRI.
• This approach offers a promising tool for clinical practice in localizing the seizure onset zone (SOZ).
• Data-driven fMRI analysis, particularly ICA, is effective even when EEG data is unreliable or unavailable.