Quantifying Interictal Spike Asymmetry on Scalp EEG to Lateralize the Seizure-Onset Zone in Temporal Lobe Epilepsy

Accurate lateralization of the seizure focus in temporal lobe epilepsy (TLE) is critical for surgical planning. Our current approach of capturing multiple seizures often misses bilateral disease because of seizure-onset variability. Interictal spike rate asymmetry is a promising marker of TLE laterality. However, visual analysis has low interrater reliability and cannot estimate long-term spike rates. This study evaluated whether asymmetry of automatically detected interictal spikes on scalp EEG can lateralize the clinician-annotated seizure focus in TLE.

We performed a retrospective cohort study of patients with drug-resistant TLE who underwent scalp EEG recording at the Hospital of the University of Pennsylvania. Spike rates were quantified using a validated deep learning model, with a novel leave-one-hemisphere-out approach developed to lateralize the side of the spikes. We developed a logistic regression model to predict the clinician-determined laterality. Model performance was assessed using the area under the ROC curve (AUC), accuracy, and area under the precision-recall curve (AUPRC).

Of 365 included patients (205 women [56.2%]; median age 38, range 18-86 years), the training cohort comprised 261 patients and the independent, time-shifted validation cohort comprised 104. In the test cohort, classification of left TLE vs right/bilateral achieved an AUC of 0.78 (95% CI 0.69-0.87), an accuracy of 70%, and an AUPRC of 0.80 (95% CI 0.69-0.91), and classification of right TLE vs left/bilateral achieved an AUC of 0.80 (95% CI 0.70-0.88), an accuracy of 75%, and an AUPRC of 0.71 (95% CI 0.56-0.85). Model-surgery concordance was positively associated with favorable surgical outcomes. Engel I outcomes were associated with higher concordance probabilities (median difference = 0.20; 95% CI -0.02 to 0.39; r = 0.35; p = 0.037), and no significant association was observed for ILAE 1-2 (median difference = 0.20; 95% CI -0.06 to 0.38; r = 0.29; p = 0.084).

Interictal spike asymmetry on scalp EEG shows promise as a noninvasive lateralizing marker in TLE. Applying this model could confirm TLE laterality when it agrees with other noninvasive data or identify patients with a high likelihood of bilateral TLE, potentially avoiding un-necessary invasive studies and reducing the number of seizures needed for evaluation.