Abstracts

Accurate Localization of Seizure Onset Zone Using Multiscale Neural Model Inversion of Resting-state Functional MRI

Abstract number : 2.323
Submission category : 5. Neuro Imaging / 5B. Functional Imaging
Year : 2024
Submission ID : 262
Source : www.aesnet.org
Presentation date : 12/8/2024 12:00:00 AM
Published date :

Authors :
Presenting Author: Guoshi Li, PhD – UNC-Chapel Hill

Varina Boerwinkle, MD – UNC
Pew-Thian Yap, PhD – UNC-Chapel Hill

Rationale: The most effective treatment for drug resistant epilepsy (DRE) is surgical rection of the seizure onset zone (SOZ), but its success critically depends on the accurate localization of SOZ. Current SOZ detection approach relies on non-invasive identification of SOZ candidates for subsequent verification with intracranial EEG (iEEG). However, with surgical failure rates between 30% to 70%, more accurate noninvasive SOZ localization is urgently needed. Resting-state function MRI (rs-fMRI) is a promising modality for accurate SOZ detection, but current statistical methods often result in multiple plausible SOZ candidates, increasing the risk for iEEG and limiting the surgical success. We aim to validate a newly developed generative modeling framework, Multiscale nEural Model Inversion (MEMI) (Fig. 1), for accurate SOZ detection. As epilepsy has long been hypothesized to result from hyperexcitablity due to imbalance between neuronal excitation and inhibition, we hypothesize that SOZ is associated with higher excitation to inhibition (E/I) ratio during interictal periods.


Methods: We applied MEMI to a rs-fMRI dataset collected from 12 DRE patients with hypothalamic hamartoma (HH) and confirmed SOZ location. MEMI estimates effective connectivity (EC) in a four-node network including the SOZ (HH) and three propagation zones (pZs) (a mesial temporal region, and left and right hippocampus). We divided the 1180 sec BOLD signals into seven overlapping sliding windows of 580 sec and a step size of 100 sec. Local and inter-regional EC was estimated in each sliding window for each subject. The regional E/I ratio was defined as ratio between total excitation strength (local recurrent excitation + all incoming inter-regional excitatory EC) and total inhibition strength (local recurrent inhibition + all incoming inter-regional inhibitory EC). We used the regional E/I ratio as a biomarker to identify SOZ.


Results: Results for two representative subjects are shown Fig. 2, demonstrating that the local excitation (WEE) and inhibition (WIE) weight of the SOZ was higher and lower than the pZs, respectively (Fig. 2A1, 2A2). This indicated a higher intra-regional E/I ratio in HH compared with other regions. Similarly, the HH also received larger inter-regional EC compared with three pZs (Fig. 2A2, B2; warmer color of the top row). Consequently, the HH exhibited the highest regional E/I ratio in most of the sliding windows (Fig. 2A3, B3). We found that for 11 out of 12 subjects, the SOZ showed a higher (mean) regional E-I ratio compared to the pZs, and for 10 out of 12 subjects, the HH displayed the largest peak E/I ratio among the four regions.


Conclusions: Our study strongly indicated that the SOZ has higher E/I ratio than the pZs during the resting-state interictal periods which can be harnessed for accurate SOZ identification. Also, MEMI is a promising approach to localize SOZ in DRE for surgical resection by estimating regional E/I ratio in the epileptic network based on rs-fMRI.



Funding: This study was funded in part by NIH grants R01EB008374 and R01MH125479 to PT-Y and R21AG083589 to GL.

Neuro Imaging