Voxel-based Automated Detection of Focal Cortical Dysplasia Lesions Using Diffusion Tensor Imaging and T2-weighted MRI Data

Yanming Wang,Yawen Zhou,Huijuan Wang,Jin Cui,Benedictor Alexander Nguchu,Xufei Zhang,Bensheng Qiu,Xiaoxiao Wang,Mingwang Zhu

DOI: https://doi.org/10.1016/j.yebeh.2018.04.005

IF: 3.337

2018-01-01

Epilepsy & Behavior

Abstract:The aim of this study was to automatically detect focal cortical dysplasia (FCD) lesions in patients with extratemporal lobe epilepsy by relying on diffusion tensor imaging (DTI) and T2-weighted magnetic resonance imaging (MRI) data. We implemented an automated classifier using voxel-based multimodal features to identify gray and white matter abnormalities of FCD in patient cohorts. In addition to the commonly used T2-weighted image intensity feature, DTI-based features were also utilized. A Gaussian processes for machine learning (GPML) classifier was tested on 12 patients with FCD (8 with histologically confirmed FCD) scanned at 1.5 T and cross-validated using a leave-one-out strategy. Moreover, we compared the multimodal GPML paradigm's performance with that of single modal GPML and classical support vector machine (SVM). Our results demonstrated that the GPML performance on DTI-based features (mean AUC = 0.63) matches with the GPML performance on T2-weighted image intensity feature (mean AUC = 0.64). More promisingly, GPML yielded significantly improved performance (mean AUC = 0.76) when applying DTI-based features to multimodal paradigm. Based on the results, it can also be clearly stated that the proposed GPML strategy performed better and is robust to unbalanced dataset contrary to SVM that performed poorly (AUC = 0.69). Therefore, the GPML paradigm using multimodal MRI data containing DTI modality has promising result towards detection of the FCD lesions and provides an effective direction for future researches.