Authors :
Presenting Author: Jared Shless, BS – Harvard Medical School
Derek Doss, BE – Vanderbilt University School of Medicine; Abhijeet Gummadavelli, MD – Clinical Fellow, Neurological Surgery, Vanderbilt University Medical Center; Aarushi Negi, BA – Vanderbilt University; Danika Paulo, MD, MSCI – Resident Physician, Neurological Surgery, Vanderbilt University Medical Center; Martin Gallagher, MD, PhD – Attending Physician, Neurology, Vanderbilt University Medical Center; Robert Naftel, MD – Attending Physician, Neurological Surgery, Vanderbilt University Medical Center; Shilpa Reddy, MD – Attending Physician, Neurology, Vanderbilt University Medical Center; Sarah Bick, MD – Attending Physician, Neurological Surgery, Vanderbilt University Medical Center; Catie Chang, PhD – Assistant Professor, Biomedical Engineering, Vanderbilt University; Victoria Morgan, PhD – Professor, Biomedical Engineering, Vanderbilt University Medical Center; Shawniqua Williams Roberson, MEng, MD – Attending Physician, Neurology, Vanderbilt University Medical Center; Graham Johnson, PhD – Medical Student, Vanderbilt University School of Medicine; Dario Englot, MD, PhD – Associate Professor, Neurological Surgery, Vanderbilt University Medical Center
Rationale:
Recent evidence suggests seizure onset zones (SOZs) in focal epilepsy are actively suppressed by local and distant brain regions during interictal periods.1,2 Specifically, SOZs exhibit high inward and low outward connectivity on stereo-electroencephalography (SEEG) analyses, which could suggest interictal suppression - this is known as the Interictal Suppression Hypothesis (ISH, Fig. 1). Yet, it is unclear how important this suppressive motif is for the global architecture of the resting state network as observable with SEEG. Thus, we employed machine learning to objectively analyze the principal SEEG network architecture. Figure 2 below provides a high level overview of this process (Figure 2). Identification of the prominent network motifs could have implications for neuromodulation targeting and understanding of focal epilepsy pathophysiology.
Methods:
To evaluate interictal SEEG network motifs, we employed principal components analysis (PCA) individually on 81 patients with focal epilepsy undergoing presurgical evaluation. Each patient had an SEEG recording obtained at Vanderbilt University Medical Center's Epilepsy Monitoring Unit (EMU). Continuous SEEG data was obtained over a 20-minute epoch in the resting state and filtered with 1-59 Hz passbands and 61-119 Hz. Based on an epileptologist's review of ictal events, SOZs, non-SOZs, and propagation zones (PZ) were assigned to each contact pair using a bipolar reference montage. For all patients, we conducted a node-based principal component analysis (PCA) of partial directed coherence (PDC), providing principal components that can be evaluated for local and global interictal electrographic motifs. Reconstruction of original PDC matrices from principal components and weights allowed for interpretation of each motif in the context of that patient’s global connectivity profile. To determine the importance of that motif in reference to SOZ connectivity, we conducted a paired t-test between the component’s weights for SOZ and non-SOZ nodes.
Results:
For 64 of 81 patients, the component that best captures ISH behavior (high inward connectivity to SOZ, and low outward connectivity to SOZ) is the first principal component (PC1). For the remaining 17 patients, the component which best resembles ISH behavior is a component other than PC1, which will be referred to as ISH PC 2+. On an electrophysiological level, this component follows ISH behavior with respect to SOZ and NIZ weight separation (Fig. 3); however, it is not the dominant network motif present in these patients. Figure 3 below delineates the explained variance for each component, by sub-cohort (Fig. 4).
Conclusions:
A high proportion of patients (79%) have an ISH-like dominant network motif that is significantly different between SOZs and Non-SOZs. The remaining 21% of patients have an ISH-like network motif, however other motifs better characterize SOZ vs. Non-SOZ interictal connectivity. It is the aim of future work to further characterize these alternate dominant motifs. Better characterization of patient-specific interictal connectivity could have implications for individual response to resective therapy and neuromodulation targeting.
Funding:
- NINDS R01NS112252, F31NS120401