Voxel-based morphometry analysis and machine learning based classification in pediatric mesial temporal lobe epilepsy with hippocampal sclerosis

Abstract

Mesial temporal lobe epilepsy with hippocampal sclerosis (MTLE-HS) is a common type of pediatric epilepsy. We sought to evaluate whether the combination of voxel-based morphometry (VBM) and support vector machine (SVM), a machine learning method, was feasible for the classification of MTLE-HS. Three-dimensional T1-weighted MRI was acquired in 37 participants including 22 with MTLE-HS (16 left, 6 right) and 15 healthy controls (HCs). VBM was used to detect the regions of gray matter volume (GMV) abnormalities. The volumes of these regions were then calculated for each participant and used as the features in SVM. The SVM model was trained and tested with leave-one-out cross validation (LOOCV). We performed VBM-based comparison and SVM-based classification between left HS (LHS) and HC as well as between right HS (RHS) and HC. Both GMV increase and reduction were found in the group comparisons with VBM. Using SVM, we reached an area under the receiver operating characteristic curve (AUC) of 0.870, 0.976 and 0.902 for the classification between LHS and HC, between RHS and HC and between HS and HC respectively. The VBM findings were concordant with the clinical findings. Thus, our proposed method combining VBM findings with SVM, were applicable in the classification of padiatric MTLE-HS with high accuracy.