Classification of Temporal Lobe Epilepsy Based on Radiomic Features of Hippocampal Subregions

To investigate whether radiomic features of hippocampal subregion can be used as biomarkers for the diagnosis of temporal lobe epilepsy (TLE) patients, and to explore the importance of some radiomics features in classification, structural magnetic resonance imaging (sMRI) was performed on 23 TLE patients and 30 healthy controls (HCs) subjects. Freesurfer 7.2 software was used for segmentation of hippocampal subregions. Radiomics features of hippocampal subregion were extracted by using 3D slicer software. After feature selection, support vector machine (SVM) was employed for the classification between TLE patients and HCs.The granule cell layer of dentate gyrus in the body of left hippocampus (GC_ML_DG-body) reached highest accuracy of 79.25%. The molecular_layer_HP in the head of right hippocampus (Molecular_layer_HP-head) has the highest classification accuracy of 79.25%. The majority of features related to classification results were second-order features, followed by first-order features and shape features. The radiomics features of hippocampal subregions can be used as a biomarker to identify TLE. The second-order features are the most important features for the classification of TLE.