In this paper, we assessed the reliability of glucose metabolic brain data for identifying lateralization of magnetic resonance imaging (MRI)-negative temporal lobe epilepsy (TLE) patients. We designed and developed an efficacious and automatic metabolic-wise lateralization framework. The proposed lateralization framework comprises three main systematic levels. In the first stage of our investigation, we pre-processed interictal fluorodeoxyglucose positron emission tomography images to extract glucose metabolic brain data. In the second stage, we used a voxel selection method involving a feature-ranking strategy to select the most discriminative metabolic voxels. Finally, we used a support vector machine followed by a 10-fold cross-validation strategy to assess the proposed lateralization framework in 27 patients with right MRI-negative TLE and 29 patients with left MRI-negative TLE. The proposed lateralization framework achieved an excellent accuracy of 96.43 % concordance with experienced PET interpreter. Thus, we show that pattern analysis of glucose metabolic brain data can accurately lateralize MRI-negative TLE patients in the clinical setting.
Keywords: Brain metabolism; Epilepsy; FDG-PET; Lateralization; Machine learning.
Copyright © 2020 Elsevier B.V. All rights reserved.