Brain MRI Segmentation for Focal Cortical Dysplasia Lesion Detection
Ivana N. Despotovic, Ief Segers, Ljiljana Platiša, Ewout Vansteenkiste, Aleksandra Pižurica, Karel Deblaere, Wilfried R. Philips · Ghent University Academic Bibliography (Ghent University) · 2011
Focal cortical dysplasia (FCD), a malformation of the cortical development in the brain, is a major cause of refractory epilepsy. In clinical treatments, FCD lesions often have to be removed by surgery and before this can be done, it is necessary to detect and delineate the lesions. However, identification of FCD lesions is a very challenging task and standard radiological MRI evaluation still fails in many cases, because of the complexity of the cortex and subtle behavior of the lesions. On T1-weighted MRI sequence, FCD is usually characterized by increased cortical thickness and blurring between cortex and white matter junction. Therefore, to automatically detect FCD lesion, it is necessary to calculate cortical thickness and intensity map, which often require accurate MRI cortex segmentation. To date, the most popular methods for brain MRI segmentation are the histogram-based method with automated threshold (HTR), Functional MRI of the Brain (FMRIB) Automated Segmentation Tool (FAST) [1] and Statistical Parametric Mapping (SPM) [2]. Although all of these techniques are fast, reproducible and require minimum human intervention, their outcome may be influenced by image noise, bias field and partial volume effects. Thus, to improve the accuracy of cortex segmentation, in this work we propose a 3D brain MRI segmentation technique based on graph cuts algorithm [3]. The novelty of our approach is automatic initialization of the voxel probabilities using Gaussian Mixture Model (GMM) and a 3D instead of 2D segmentation using three labels: gray matter, white matter and cerebrospinal fluid (CSF).