Abstracts

Rationale: Focal cortical dysplasia (FCD) is found in approximately one-half of patients with medically refractory epilepsy. The common features of FCD are blurring of the gray-white matter junction, cortical thickening, abnormal gyral and sulcal contours, and increased signal intensity on T2-weighted images (Huppertz, et al. 2005). However, sometime the diagnosis of burring of the gray-white matter junction is time-consuming and unrevealing in high percentage of patients. Hence, this research focused on enhancement of blurring of the gray-white matter junction and gray-white matter segmentation. Methods: Several artificial neural network (ANN) structures were used for brain segmentation in the research, such as C-means clustering (CM), fuzzy C-means clustering (FCM), improved fuzzy C-means (IFCM), and decision-based neural networks (DBNN). These methods were applied to segment a MRI brain image into gray matter, white matter, and cerebral spinal fluid (CSF). The accuracy of segmentation was also evaluated by 50 MRI images from 10 different individuals (5 images for each person) by using IBSR( http://www.cma.mgh.harvard.edu/ibsr/ ) database. Nevertheless, the blurring area of the gray-white matter junction is the potential FCD area. Our goal is to enhance those suspect areas for FCD pre-screening by using so-called fuzzy analysis. First, the method located the FCM fuzzy points with index values around 0.5 which means those points have 50% chance to classify into gray matter and the other 50% chance to classify into write matter. That shows the blurring areas. After the fuzzy points are selected, the fuzzy indicator matrix of the FCM method combined with 2D Fourier filtering to demonstrate the enhanced gray-white matter junction. Results: For brain image segmentation, our results showed incorrect segmentation (InC) rates are 0.15343, 0.2091, 0.29899, and 0.30587, for DBNN, CM, IFCM, and FCM methods respectively. That is, DBNN method is more accurate than other methods. However, CM method is more reliable if the intensity of MRI image is not calibrated. The fuzzy indicator matrix of FCM method can show the degree of blurring of the gray-white matter junction. It combines with a 2D Fourier filter to enhance FCD area for helping physicians to diagnosis the burring area. Figure 1 showed that the FCD area is successfully identified after image enhancement. The results of the case also been confirmed by EEG signals and our lab reports.Conclusions: Our methods evaluated the different methods for brain segmentation. Based FCM segmentation technologies, we enhanced the blurring of the gray-white matter junction area, which may not easily to be observed, without using normal brain database (NDB).