Automatic Segmentation of Brain Tumors on Non-Contrast-Enhanced Magnetic Resonance Images using Fuzzy Clustering

Manual brain tumor segmentation from magnetic resonance imaging is a difficult and time-consuming task for physicians. For this reason, an automated brain tumor segmentation method is desirable. Currently, segmentation of gadolinium-enhanced tumor is feasible via combining semi-supervised clustering with knowledge-based analysis [1]. However, the accuracy of supervised segmentation techniques depends on the performance of human experts. Moreover, approximately 10% of all brain tumors can not be enhanced in magnetic resonance (MR) images after a contrast agent has been administered, making segmentation extremely difficult [2]. Therefore, we define a non-contrast-enhanced brain tumor segmentation with an unsupervised clustering algorithm, Fuzzy C-Means (FCM) and there is no need for manual data labeling. There are tumor segmentation studies based on non-contrast-enhanced T1-weighed (T1), T2-weighed (T2) and proton density weighed (PD) images [2, 3]. Our study is based on T1 and T2 images only. Even though FCM is quite popular, the neighborhood condition is not considered in FCM. Region merging used in this study just makes up for that weakness. The purpose of this study is to segment brain tumor with non-contrast-enhanced MR imaging via unsupervised FCM clustering combined with region merging and knowledge-based analysis. Material and Methods