Authors :
Presenting Author: Georgios Ntolkeras, MD – Boston Childrens Hospital
Vitor Lauar Pimenta de Figueiredo, MS – Faculty of Medicine of the University of São Paulo
Navaneethakrishna Makaram, PhD, MS – Boston Children's Hospital, Harvard Medical School, Boston, MA 02115, USA
Ellen Grant, MD – Boston Childrens Hospital
Scellig Stone, MD, PhD – Boston Childrens Hospital & Harvard Medical School
Phillip Pearl, MD – Boston Children's Hospital & Harvard Medical School
Alexander Rotenberg, MD, PhD – Boston Children's Hospital, Harvard Medical School, Boston, MA, USA
Eleonora Tamilia, PhD – Boston Children’s Hospital, Harvard Medical School, Boston, MA, USA
Rationale:
Corpus callosotomy (CC) is a palliative surgical procedure for drug-resistant epilepsy (DRE). Seizure reduction after CC is difficult to predict, with 45% of patients exposed to the potential adverse effects of this major procedure without gaining significant clinical benefits. We examined the prognostic value of deconstructing presurgical brain networks derived from routinely collected scalp EEG prior to CC, for purposes of forecasting the liklelihood of meaningful seizure reduction. We aim to establish EEG-derived functional connectivity (FC) metrics as presurgical predictors of CC outcomes and build an EEG-based machine learning algorithm to ultimately improve patient selection for CC.
Methods:
We analyzed 5 minutes of interictal scalp EEG data from 32 children with DRE who underwent CC (median age: 10.5 years, median time from EEG to surgery: 4.8 months). We estimated the brain FC in five frequency bands (delta, theta, beta, alpha, gamma), (Fig. 1A) and computed the following FC metrics: whole brain FC (WB-FC), within-hemisphere FC (WH-FC), between-hemisphere FC (BH-FC), and interhemispheric asymmetry (IHA), (Fig. 1B). We classified each patient into three outcome categories based on their seizure reduction (Excellent: >90%, Intermediate: 50-90%, Poor: < 50%) (Fig 1C). We then examined the correlation of each presurgical FC metric with the patients’ seizure outcome category (logistic regression). The effect of clinical characteristics on the outcome was also similarly tested. We also compared pre- and post-surgical FC (Wilcoxon signed-rank test) separately in optimal (excellent) and suboptimal (intermediate and poor) outcome groups. Finally, we developed a multi-feature classification model (support-vector-machine, SVM) to predict outcome (leave-one-out cross-validation).