Authors :
Presenting Author: Derek Doss, PhD – Vanderbilt University
Ghassan Makhoul, BS – Vanderbilt University Medical Center
Graham Johnson, MD, PhD – Mayo Clinic
Anas Reda, MS – Vanderbilt University Medical Center
Bruno Hidalgo, BS – Vanderbilt University Medical Center
Addison Cavender, BS – Vanderbilt University Medical Center
Emily Liao, BE – University of Minnesota
Kate Wang, BE – Vanderbilt University Medical Center
Sameer Sundrani, – Vanderbilt University Medical Center
Haatef Pourmotabbed, MS – Vanderbilt University
Sarah Goodale, PhD – Vanderbilt University
Douglas Terry, PhD – Vanderbilt University Medical Center
Hakmook Kang, PhD – Vanderbilt University Medical Center
Kevin Haas, MD, PhD – Vanderbilt University Medical Center
Martin Gallagher, MD, PhD – Vanderbilt University Medical Center
Victoria Morgan, PhD – Vanderbilt University Medical Center
Catie Chang, PhD – Vanderbilt University
Dario Englot, MD PhD – Vanderbilt University Medical Center
Rationale:
In addition to recurrent seizures, patients with focal epilepsy can experience neurocognitive deficits, including decreased attention, executive function, and concentration. These neurocognitive can greatly affect the quality of life for many patients, but it is not fully understood why these neurocognitive deficits occur. The Extended Network Inhibition Hypothesis (ENIH) posited that impaired subcortical-to-neocortical arousal networks may contribute to neurocognitive deficits. While prior studies have demonstrated abnormalities in subcortical-to-neocortical functional connectivity (FC), overall arousal level estimates have been similar between patients with focal epilepsy and healthy controls. Thus, we sought to investigate the low arousal state in a data driven manner to identify possible widespread FC differences in the low arousal state in patients with focal epilepsy that may contribute to neurocognitive deficits.
Methods:
20-minutes of resting-state functional MRI (fMRI) data were obtained for 103 healthy controls and 82 patients with focal epilepsy. These data were parcellated with the Harvard-Oxford cortical atlas, the Harvard-Oxford Ascending Arousal Network atlas, a patient specific thalamic atlas, and a patient-specific atlas of nucleus basalis of Meynert (NBM). These data were preprocessed as described in prior studies and dynamic FC was computed using 1-minute windows with a 30-second stride. Deep-learning clustering was performed to identify brain states in both patients and controls separately using Variational Autoencoders (VAEs) and k-means clustering. Arousal level of each identified state was computed using an fMRI-based arousal estimator. The lowest arousal FC state was analyzed for both patients and controls. The percentage of time each subject spent in the lowest arousal state was computed. Segregation, a measure of reduced between and increased within network FC was computed for the low arousal FC state in each subject.
Results:
We found that the low arousal FC states qualitatively appeared different between patients and controls (Fig. 1A), with networks showing less structure in patients vs controls. Furthermore, it was found that patients spent more time in the low arousal state than controls (Fig. 1B, p=6.89E-6, two-sample t-test). Finally, it was found that the low arousal FC state in controls had a higher segregation than patients (Fig. 1C, p=1.94E-6). Considering that physiologically appropriate low arousal states are associated with a high segregation, this result may represent a pathological low arousal state in patients with epilepsy.
Conclusions:
Neurocognitive deficits can be devastating to patients with epilepsy, and it is not fully understood why these widespread deficits occur. In this work, we use a data-driven deep-learning method to identify FC states associated with low arousal and demonstrate that low arousal FC states are more common in patients with epilepsy and may represent pathological low arousal state that could contribute to the global neurocognitive deficits in patients with epilepsy.
Funding:
This work was funded by NIH grants T32EB021937, T32GM007347, F31NS120401, R01NS112252, R0INS110130, and R01NS10844.