Unsupervised learning algorithms for MRI brain tumor segmentation

Abstract

The structure of the brain can be seen by the Magnetic Resonance (MR) image output. MR scanned image of the brain is utilized for the entire study in this paper. The MR image filter is more agreeable than some other outputs for analysis. It will not influence the human body since it does not hone any radiation. In digitization of MR scanned image, segmentation of brain tumor is one kind of challenging problems and it is critical to clinical diagnosis. So segmentation needs to be accurate, robust, and efficient to avoid impacts caused by various large and complex biases added to images. Clustering algorithms have been widely used for the segmentation. In this paper, the K-means (KM) clustering and Fuzzy C-means (FCM) clustering algorithms are used to locate the tumor and extract it. Comparative analysis in terms of Segmented area, Relative area, Mean Squared Error (MSE) and Peak Signal to Noise Ratio (PSNR) is performed between K-means clustering and FCM clustering algorithms. The obtained performance measures from the experiments indicate the superiority of the chosen FCM algorithm over the K-means algorithm. That is 0.93% of relative segmented tumor area for FCM shows that the area which was effected by the tumor in the original MR image is segmented as a tumor. The FCM Algorithm has less processing time of 8.639 seconds compared to 22.831 seconds for KM algorithm.