Abstracts

Rationale: Given the heterogeneity of epilepsy, the need to move beyond a “one-size-fits-all" therapeutic approach for seizure control and associated cognitive and neuropsychiatric comorbidities remains a top priority. With continuing technological advancements, we can expand the field of precision medicine in epilepsy beyond defining causative genetic mechanisms to include identifying and characterizing abnormal neural networks. One of the emerging tools for measuring the macro-scale connectivity abnormalities present in epilepsy is high-density scalp EEG (hdEEG). However, the optimal approach to quantify EEG connectivity remains unclear. Topological Data Analysis (TDA) is a well-developed mathematical discipline with utility in network science. The potential benefits of this method over the more commonly implemented graph theory analysis include an intrinsic solution to the problem of thresholds versus weighted edges in adjacency matrices called filtration and the ability to move beyond dyadic structure of nodes and edges to higher dimensional relationships.

Methods: Following IRB approval and subject recruitment, we compared EEG connectivity from a total of 12 patients with focal epilepsy to 11 age-matched controls. Correlation matrices of each subject were created from 6-sec artifact-free and epileptiform discharges-free eyes-open resting state segments by performing parcellation using the Destrieux atlas in Brainstorm with the Brainstorm mean Pearson correlation function. These matrices were then made into distance matrices, processed into Vietorsi-Rips Complexes to calculate persistent homology, and eventually into landscape functions for dimensions 0, 1,and 2 for group comparison. Energy statistics were used to compare the mean landscape persistence between epilepsy and control groups. Machine-learning with a support vector machine (SVM) with a linear kernel was used to assess the ability to detect epilepsy using hdEEG derived Persistent Homology.

Results: Epilepsy patients had greater Persistent Homology in dimension 1 (p=0.009) and 2 (p=0.03), but not 0 (p=0.42) (Fig 1) p-values were corrected for false discovery rate. SVM and leave-one-out-cross-validation, Persistent Homology features achieved a 0.87 area-under-the-ROC curve (Fig 2).

Conclusions: These preliminary results indicated that hdEEG connectivity analyzed with TDA is a promising method to differentiate epilepsy from controls. These findings demonstrate that there is increased higher dimensional connectivity in epilepsy patients, but not in the lower dimension of 0 which is equivalent to “nodes” in graph theory. These results are supportive of tightly interconnected regions that form the holes and voids analyzed with TDA suggestive of an “epileptic-network” that may have complex higher dimensional relationships. Implications of this study include a potential novel method for identifying diagnostic and prognostic biomarkers of aberrant connectivity in epilepsy.

Funding: Please list any funding that was received in support of this abstract.: University of Wisconsin Institute for Clinical and Translational Research; NIH-NINDS R01-NS111022.