DEPICT THE EPILEPTOGENIC REGION ON BRAIN MRI BY USING FUZZY C-MEANS INDEX MATRIX
Abstract number :
3.185
Submission category :
5. Human Imaging
Year :
2008
Submission ID :
9316
Source :
www.aesnet.org
Presentation date :
12/5/2008 12:00:00 AM
Published date :
Dec 4, 2008, 06:00 AM
Authors :
Tsu-Wang Shen and Yue-Loong Hsin
Rationale: Focal cortical dysplasia (FCD), an increasingly recognized cause of intractable epilepsy, is often associated with blurring of the gray-white matter junction in magnetic resonance images (MRI). MRI constructs the image by applying magnetic field to align the nuclear magnetization of hydrogen atoms in water in the body. It is also well-known that tissue property changes associated with FCD lead to a reduction in free water and resultant shortening of T2 and T1 relaxation times. However, at the early stage of FCD, tissue structure changes may be trivial, especially for T1 images, which may not be visible by eyes. Recent report from Yoshida (2007) indicates a patient with intractable epilepsy and negative MRI at the age of 2.5 years. And the follow-up MRI revealed typical FCD findings in the right frontal lobe. Hence, we aimed to develop an analysis system to recognize the focal cortical lesion that is undetectable visually. Methods: Initially, seven subjects were investigated in this research for a double-blind experiment. All subjects had defined ictal region/s that localized by long-term video EEG. They had no abnormal MRI features found by visualization according to neuroradiologists. Three patients had focal cortical resection. In addition, the algorithm developer did not know any information about those subjects in advance, such as FCD location and size, abnormal EEG patterns and places as well. Magnetic resonance imaging was performed on a 1.5-T scanner (GE medical systems) with a standard circular polarized head coil. Fuzzy C-means (FCM) is a widely-used unsupervised learning method in artificial neural networks for classification. The method modifies the traditional C-means indicator into a fuzzy indicator, so the clustering can be performed based on (a) the distances between each point and codewords; and (b) the degrees of membership for each point. For FCM, the power index m is equal to 2. Then, the fuzzy indicator matrix I with threshold at 0.6 can show the degree of blurring of the gray-white matter junction. Hence, I= all element of I (xn, i) <0.6. The fuzzy indicator matrix I (512×512) is divided into 64×64 pairs of square fragments. Assume that a total number of 8×8 pixels in one fragment. The figure 1 shows the original MRI image, fuzzy indicator matrix I, and square fragments. It can be easily observed that the intensity of sum of those fuzzy points is very high in abnormal area by comparing with image background. Results: Without any visualization abnormal from MRI, our double-blind study successfully determined 5 out of 8 FCD cases which were verified by interictal EEG signals. The details for the clinical data and MRI analysis results are listed on table 1. Conclusions: The double-blind study showed that our method can potentially be used as an automated MRI prescreening tool, even there is no significant to be found by visualization. That may help doctor to identify potential FCD patients with intractable epilepsy and negative MRI evaluation.
Neuroimaging