Abstracts

Rationale:
Epilepsy impacts ~50 million people globally, but not all people’s seizures are responsive to medication. Surgical intervention may reduce disease burden but requires epileptogenic zone(s) localization. Stereotactic electroencephalography (SEEG) is a minimally invasive technique where electrodes monitor brain activity. Despite SEEG being the gold standard for epileptic zone localization, it is limited by sampling bias. Functional magnetic resonance imaging (fMRI) samples the entire brain through noninvasive imaging but has had limited success in identifying epileptogenic zone(s). This investigation tests if a model relating fMRI to SEEG measures of connectivity, could enable predictions of epileptogenic zone(s) from noninvasive fMRI.
Method:
For 20 patients with focal epilepsy (16 females, 34.3±10.8 years), we acquired 20 minutes of resting-state fMRI and 2 minutes of resting-state SEEG. FMRI resting-state was collected for 69 healthy controls (36 females, 39.0±12.4 years). Using clinical practice, SEEG regions, designated by Desikan-Killiany (DK) atlas, were classified as epileptogenic or non-epileptogenic. Between connectivity refers to mean connectivity between a region and remaining brain regions sampled. From SEEG, between, alpha band imaginary coherence was calculated and z-scored. For fMRI, Pearson correlation, was calculated between 89 patient-specific, DK regions. Patient fMRI correlation values were corrected by controls, by subtracting age based linear fits in all connections. Corrected correlation matrices were reduced to match SEEG regions, and between Pearson correlation was calculated. FMRI between connectivity was rescaled per patient and fit with a quadratic function to SEEG between connectivity, in order to have the fMRI metric informed by the SEEG between connectivity.
Results:
SEEG between connectivity in epileptogenic regions was higher than non-epileptogenic regions (p< 0.01, t-test) (Fig.1A). However, using fMRI between connectivity, age corrected by controls, was not different between categories (p >0.05, t-test) (Fig.1B). After rescaling and model fitting fMRI to SEEG between connectivity, rescaled and fit values were higher in epileptogenic regions (p< 0.05, t-test) (Fig.1C). Using a receiver operator characteristic curve to quantify ability to predict epileptogenic zones versus non-epileptogenic zones, model fit, rescaled fMRI between connectivity was better able to predict epileptogenic regions than fMRI between connectivity. Similar trends were observed when fMRI resting-state was split into two 10 minute segments and tested independently.
Conclusion:
Unlike baseline fMRI connectivity metrics, our model fit fMRI between connectivity was able distinguish between epileptogenic and non-epileptogenic zones. The ability of model fit fMRI to identify epileptogenic zones now resembled that of SEEG. If validated in a larger cohort, fMRI metrics model fit to SEEG metrics, may have the potential to noninvasively localize epileptogenic zone(s).
Funding:
:NIH T32 EB001628-17, F31 NS106735, R00NS097618, T32GM734740, T32 EB021937, R01 NS112252.