Authors :
Presenting Author: Olivia Lee, BA – Cincinnati Children's Hospital Medical Center
Seungrok Hong, BS – Cincinnati Children's Hospital Medical Center; Paul Horn, PhD – Statistician, Pediatric Neurology, Cincinnati Children's Hospital Medical Center; Hisako Fujiwara, PhD – Lab Manager, Cincinnati Children's Hospital Medical Center; Jesse Skoch, MD – Pediatric Neurosurgery – Cincinnati Children's Hospital Medical Center; Ravinda Arya, MD – Cincinnati Children's Hospital Medical Center; Gewalin Aungaroon, MD – Cincinnati Children's Hospital Medical Center; Susan Fong, MD, PhD – Cincinnati Children's Hospital Medical Center; Katherine Holland-Bouley, MD, PhD – Cincinnati Children's Hospital Medical Center; Kelly Kremer, MD – Cincinnati Children's Hospital Medical Center; Katie Ihnen, MD, PhD – Cincinnati Children's Hospital Medical Center; Nan Lin, MD – Cincinnati Children's Hospital Medical Center; Wei Liu, MD – Cincinnati Children's Hospital Medical Center; Heather Wied, MD, PhD – Cincinnati Children's Hospital Medical Center; Francesco Mangano, DO, FACS, FAAP, FACOS – Pediatric Neurosurgery – Cincinnati Children's Hospital Medical Center; Hansel Greiner, MD – Cincinnati Children's Hospital Medical Center; Jeffrey Tenney, MD, PhD – Principal Investigator, Pediatric Neurology, Cincinnati Children's Hospital Medical Center
Rationale:
Temporal Lobe Epilepsy (TLE) is the most prevalent epilepsy syndrome, yet about forty four percent of pediatric patients who undergo resective surgery fail to achieve seizure freedom. Recent research indicates that surgical failure may arise from epileptogenic networks extending beyond temporal regions, termed temporal plus epilepsy (TLE+). Differentiating between TLE and TLE+ is vital, as the approach for planning intracranial EEG (iEEG) and surgical resection hinges on this distinction. Magnetoencephalography (MEG) is a pre-surgical tool, facilitating the determination of targets for subsequent iEEG monitoring. This study utilizes MEG data to refine the classification of temporal seizure subtypes, potentially enhancing surgical outcomes.
Methods:
This retrospective analysis involved eighty patients who underwent both pre-surgical MEG and subsequent iEEG monitoring for medication resistant epilepsy. User-defined virtual sensors (UDvs) were symmetrically placed in predefined anatomical locations encompassing mesial temporal structures (amygdalohippocampus), inferior/middle/superior temporal gyri, and neighboring structures (insula, suprasylvian operculum, orbitofrontal cortex, TPO junction) at 10mm spacing. Three ten-minute MEG recordings per patient were examined to classify the presence or absence of epileptiform activity in each predefined anatomical location. Employing k-modes cluster analysis, similarity between UDvs-MEG patterns was assessed. Bootstrapping with a gap-statistic determined the optimal cluster count to be nine clusters. Fisher’s Exact Test for Count Data and individual pairwise t-tests correlated the International League Against Epilepsy (ILAE) outcome measures 1-year post-resection with anatomical subgroup clusters.
Results:
After the number of clusters were determined, k-modes clustering determined the nine clusters of MEG spikes to be 1) hippocampus, insula 2) all areas 3) hippocampus, temporal, orbitofrontal 4) temporal 5) hippocampus, temporal 6) hippocampus, temporal, insula, orbitofrontal 7) hippocampus, temporal, insula, and operculum 8) hippocampus 9) hippocampus, temporal, insula. Fisher’s Exact Test for Count Data indicated that the clusters are significantly correlated with seizure freedom one year post-resection (p-value = 0.000813). Pairwise t-tests found clusters #1(hippocampus, insula, p-value = 0.03125) and #7 (hippocampus, temporal, insula, and operculum, p-value = 0.02246) to be significantly correlated with worsened seizure outcome one year post-resection.
Conclusions:
This study provides evidence that specific subtypes of temporal epilepsy can be identified using a data-driven clustering approach. Additionally, some of the epileptogenic subtypes found are associated with poorer surgical outcomes. In the future, we intend to expand the number of participants in the study and include iEEG data in the analysis of post-surgical outcomes. Ultimately, we hope to create a machine learning algorithm to more accurately plan individual iEEG placement based on noninvasive MEG testing, improving surgical resection outcomes.
Funding:
University of Cincinnati College of Medicine, Child and Adolescent Medical School Scholars Program