Abstracts

CRYPTOGENIC (MRI-NEGATIVE) EPILEPSY: AUTOMATIC SEIZURE FOCUS LATERALIZATION

Abstract number : 1.110
Submission category : 4. Clinical Epilepsy
Year : 2012
Submission ID : 15585
Source : www.aesnet.org
Presentation date : 11/30/2012 12:00:00 AM
Published date : Sep 6, 2012, 12:16 PM

Authors :
S. Hong, H. Kim, D. Scharder, N. Bernasconi, A. Bernasconi

Rationale: MRI has revolutionized the diagnosis of drug-resistant epilepsy by allowing the detection of lesions associated with the epileptogenic zone, thus leading to increased rates of successful resective surgery. However, in many patients, best-practice MRI is unable to reveal a surgical target [1]. In these MRI-negative (or cryptogenic) cases diverse seizure semiology and non-localizing EEG findings make the formulation of an a priori hypothesis about the epileptogenic zone often very difficult [2]. Indeed, surgery in the absence of an MRI-visible lesion is one of the greatest clinical challenges in many tertiary centers. There is need for alternative approaches to indentify the epileptogenic focus in these patients. We previously showed that cryptogenic epilepsy is associated with widespread bilateral cortical atrophy involving particularly the frontal lobes [3]. The purpose of this study was to automatically lateralize the seizure focus using MRI-based analysis of cortical thickness. Methods: We studied 37 patients (22 males, 30±8 years) with cryptogenic, drug-resistant extra-temporal lobe epilepsy. The focus was pre-central (frontal or fronto-central) in 22 and post-central (centro-parietal, parietal or parieto-occipital) in 15 (left-sided in 21 and right-sided in 16). In each region of atrophy identified by our previous analysis [3] we calculated mean cortical thickness and z-scores relative to the distribution of 41 age- and sex-matched healthy controls (16 males, 31±11 years) on 1.5T MRI (3D T1-fast field echo sequence; TR=18 ms; TE=10 ms, voxel size=1mm3). We then fed into a linear discriminant classifier each cluster's mean thickness and the mean value of the contralateral homologous region. To avoid biases, we used iteratively various significance levels ranging from α=0.025 to α=0.0001. We estimated the performance of our predictive model by performing a leave-one-out cross-validation (each patient is lateralized using data of all other subjects). Results: We correctly lateralized the focus in 77% of patients using the fronto-opercular cluster of atrophy, in 74% using the occipital cluster, and in 68% using the fronto-polar and fronto-central clusters. The frontal clusters yielded a correct classification in a higher proportion of patients with pre- than post-central foci (fronto-polar and fronto-central: 90% vs. 46%, p<0.05; fronto-opercular: 81% vs. 67%). Conversely, the occipital cluster lateralized accurately post-central foci (86% vs. 68%). Combining all clusters, we lateralized 86.5% of patients (pre-central: 86%; post-central: 87%). Conclusions: Our fully automated classifier enabled accurate and reliable lateralization of the seizure focus in the majority of patients with unremarkable conventional MRI. Higher accuracy obtained when using clusters nearby the putative seizure focus implies that, despite widespread changes, atrophy may be a biomarker for epileptogenicity. This method may substantially improve MRI analysis so that it can fulfill its role for surgical target localization in cryptogenic epilepsy.
Clinical Epilepsy