Identification of Negative BOLD Responses in Epilepsy Using Windkessel Models.

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Epilepsy patients exhibit positive and negative BOLD responses (NBRs) during interictal epileptic discharges.
• Four potential mechanisms underlie NBRs: neuronal disruption, altered neurometabolic/vascular coupling, arterial blood stealing, and enhanced cortical inhibition.
• Accurate classification of NBR mechanisms is crucial for improving electroencephalography-functional magnetic resonance imaging (EEG-fMRI) specificity in identifying seizure-onset zones in refractory epilepsy.

Purpose of the Study:

• To develop and validate a model for classifying the four proposed NBR mechanisms based on their hemodynamic response functions (HRFs).
• To investigate the BOLD signal fingerprints of different NBR mechanisms using a biophysical model.
• To assess the efficacy of machine learning in classifying these mechanisms from simulated and real EEG-fMRI data.

Main Methods:

• Utilized a Windkessel model with viscoelastic properties, coupled with dynamic models of neuronal activity and tissue/blood oxygenation.
• Evaluated the impact of key model parameters on BOLD responses.
• Employed a general linear model to represent NBRs and a machine learning classifier trained on simulated HRFs for mechanism prediction.

Main Results:

• The study demonstrated that a general linear model can accurately represent the four distinct NBR types.
• A machine learning classifier, trained on simulated HRFs, successfully classified the four mechanisms from realistic fMRI BOLD signals via cross-validation.
• Application to EEG-fMRI data from five epilepsy patients suggested the presence of these mechanisms.

Conclusions:

• Biophysically inspired models combined with general linear models can accurately identify and interpret NBR mechanisms in epilepsy.
• This approach holds promise for enhancing the diagnostic capabilities of EEG-fMRI for refractory epilepsy.
• Understanding NBR mechanisms can lead to more precise localization of seizure-onset zones.