Abstracts

Rationale: To identify how temporal lobe epilepsy (TLE) affects functional networks and regions beyond the temporal lobe, we used a machine learning approach for comparing fMRI time series data of patients to controls, and generated group functional anomaly maps (FAM, NeuroImage 2020;215;116806). We compared different parcellation atlases using this method and also mapped the regions that distinguish left versus right TLE.

Methods: Resting state fMRI images from 46 TLE patients (14 left and 32 right) and 94 controls from a single center were included. Subjects had two 10-minute runs performed in a 3T scanner in awake state with eyes closed (voxel size 3x3x4, repetition time 2s). All images were registered to a T1-weighted image and resampled into a standard MNI space. Preprocessing was performed using fMRIprep. Four atlases were used to generate 100, 200, 400 cortical (Schaefer functional atlases), or 112 cortical and subcortical (Harvard-Oxford anatomical atlas) brain parcels. Denoised time series were extracted and averaged within each parcel using xcpEngine pipeline.

To generate FAMs for left and right TLE, two separate support vector regression models were estimated by comparing each TLE group’s fMRI time series to those of controls. In a separate model, left and right TLE groups were compared, without controls, to create a distinctive FAM. Each parcel was given a weight based on its contribution to the model, and parcels were ranked for each atlas based on their importance (absolute value of their weights). Relative importance of networks/regions to the models were evaluated by their representation in the 20 most important parcels on this ranking. The weight values were projected back to the standard brain to generate left or right TLE FAMs.

Results: For both right and left TLE, our method showed distributed changes in fMRI time series beyond the temporal regions. Relative to all regions/networks, there was a higher representation of parcels from the visual and somatosensory networks in functional Schaefer atlases, and parahippocampal, hippocampal, and amygdala parcels in Harvard-Oxford atlas.

When comparing left to right TLE patients without controls, the distinctive FAM showed higher contribution of parcels in the inferior and anterior right temporal lobe region and frontoparietal control network distinguishing the two TLE groups, relative to other regions (figure).

Conclusions: Functional anomaly maps reveal extensive changes in TLE compared to healthy controls. Primary visual and sensorimotor, as well as mesial temporal regions appear to be more sensitive in highlighting these changes based on our dataset. Distinctive FAM patterns in comparing right and left TLE patients are consistent with prior clinical and functional connectivity studies. Evaluation of spatiotemporal functional changes using this method can add to functional connectivity approaches for investigating the effect of epileptic network on brain function. Future studies will examine whether this method can be applied to single patients for disease classification.

Funding: Please list any funding that was received in support of this abstract.: TG:NIH/NCATS UL1TR001876/KL2TR001877;VM:NIH NS075270,NS110130,NS108445; DE:NS097618;AD:K12HD093427.