Computer Aided Diagnosis of Intractable Epilepsy with MRI Imaging Based on Textural Information

• Published in: 2013 International Workshop on Pattern Recognition in Neuroimaging pp. 90 - 93
• Main Authors: El Azami, Meriem, Hammers, Alexander, Costes, Nicolas, Lartizien, Carole
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2013
• Subjects: Epilepsy; Feature extraction; Focal Cortical Dysplasia; Intractable epilepsy; Junctions; Kernel; Magnetic resonance imaging; MRI; One-class SVM; Single subject analysis; SPM; Support vector machines
• DOI: 10.1109/PRNI.2013.32

We designed a machine learning system based on a one-class support vector machine (OC-SVM) classifier in view of the detection of abnormalities in magnetic resonance images (MRIs) of patients with intractable epilepsy. This system performs a voxelwise analysis and outputs clusters of detected voxels ranked by size and suspicion degree. Features correspond to a combination of six maps: three tissue probabilities (grey matter, white matter and cerebrospinal fluid), cortical thickness, grey matter extension, and greywhite matter junction. The OC-SVM is trained using 29 controls, and tested on two patients with histologically proven focal cortical dysplasia (FCD). To assess the performance of the OC-SVM classifier, the classifier was compared with a statistical parametric mapping (SPM) single subject analysis using junction and extension maps only. The identified regions were also visually evaluated by an expert and compared to other data such as FDG-positron Emission tomography (PET) and magneto encephalography (MEG). For the two patients, both analyses agreed with the visually determined localization of the FCD lesions. No match was found for the other detected regions. The OC-SVM classifier was more specific in region localization and generated fewer false positive detections than the mass-univariate SPM approach.