(K-420) Structural MRI to Quantify Brain Networks in Epilepsy

Sub-Track: Neuroimaging

Diffusion-Weighted Imaging (DWI) is an MRI modality that identifies white matter tracts in the brain based on the diffusion of water in response to a varying magnetic field. DWI enables the identification of epileptic networks-regions in the brain responsible for the origination and propagation of seizures in epilepsy patients. However, geometric distortions result from magnetic field inhomogeneities (tissues in the brain exhibit varying susceptibility to a magnetic field), the motion of the subject's head, and eddy currents which result from changing the magnetic field direction to observe white matter tracts combined with the fast EPI (echo planar imaging) sequence. To correct these distortions, field maps and reverse phase-encoded images are utilized. Although beneficial, this neuroimaging data is not commonly a part of the scanning protocol in the clinical setting, consequently hindering the reliability of DWI in delineating epileptic networks.

This project investigated three methods of preprocessing DWI data:  Positive and Reverse Phase-Encoding (Blip-Up Blip-Down), Phase Magnitude (Phase and Magnitude images generated field maps), and Synthetic Field Maps. FSL Topup generated field maps and corrected for magnetic field inhomogeneities of the non-diffusion-weighted image, and FSL Eddy corrected for eddy distortions. The Synbo-Disco model, a publicly available deep learning model, generated an undistorted b0 image without a reverse phase-encoded image by estimating the anatomy of the corrected b0 DWI from a T1W structural image.  The distortion-corrected diffusion images were fiber-tracked, and white matter tracts were filtered by ROI. Next, connectivity matrices were generated and metrics characterizing the connections between brain areas were quantified: connectivity track count, mean and median fractional anisotropy (FA), mean and median track length, mean and median mean diffusivity (MD), mean and median axial diffusivity (AD), mean and median radial diffusivity (RD), and mean and median quantitative anisotropy (QA).  These connectivity metrics were used to determine the Pearson and Spearman correlations between all three preprocessing methods for brain networks of subjects with and without epilepsy.  Correlations were also separated by lobe to determine if there are regions of the brain that exhibit a higher correlation of the connectivity matrices between preprocessing methods. 

In this project, several metrics describing the connectivity between brain parcellations in networks of patients with and without epilepsy were quantified, with the highest correlation demonstrated for Connectivity Track Count. The Temporal, Frontal, and Parietal Lobes exhibited the highest values for the Track Count correlations (Pearson and Spearman) between the Phase Magnitude and Blip-Up Blip-Down DWI preprocessing methods separated by lobe, possibly due to these regions being epileptogenic in subjects experiencing seizures. It is also important to note that connectivity metrics for the epilepsy subject exhibited higher correlations than the non-epilepsy subject. Furthermore, the high correlations in all quantitative metrics of connectivity between Synthetic Field Maps and the advanced preprocessing methods Blip-Up Blip-Down and Phase Magnitude demonstrate that AI generated field maps present an opportunity for preprocessing in the clinical setting when field maps and reverse phase-encoding are not available.

Thank you, Dr. Kathryn A. Davis and Dr. Nishant Sinha, for their mentorship and guidance in this research opportunity. The work was supported by Penn Engineering’s Rachleff Scholars Program as part of the 2023 summer research experience and the research was conducted at the Center for Neuroengineering and Therapeutics.

[1] Sinha et al. 2022, Epilepsy Current,
https://doi.org/10.1177/15357597221101271 

 [2] Sinha et al. 2021, Neurology, https://doi.org/10.1212/WNL.0000000000011315

[3] Schilling, Kurt G., et al. “Distortion Correction of Diffusion Weighted MRI without Reverse Phase-Encoding Scans or Field-Maps.” PLOS ONE, vol. 15, no. 7, 2020, https://doi.org/10.1371/journal.pone.0236418. Github: https://github.com/MASILab/Synb0-DISCO