Automated Brain Tumor Detection Using Discriminative Clustering Based MRI Segmentation

This paper presents a framework for detecting a tumor from a brain MR image automatically using discriminative clustering based brain MRI segmentation. The main objective of this paper is to perform an automatic brain tumor detection which uses superpixel zoning for its initial segmentation, which reduces computational overhead and uses discriminative clustering which accounts for tissue heterogeneity in brain MR images. In the past few years, automated brain tumor detection has become an effective topic of research in medical diagnostics and clinical expedition. Superpixel zoning of brain tissues from a brain MR image is used in this paper and superpixel zones are constructed by analyzing the intensity values of the neighborhood pixels. In clustering of these brain tissues, it still faces challenges such as tissue heterogeneity and redundancy of MRI features. To encounter these challenges, we have used a discriminative clustering method to segregate the vital regions of brain such as cerebro spinal fluid (CSF), white matter (WM) and gray matter (GM). This method uses a haar wavelet transform, which generates candidate area matrix vectors. These vectors are transformed into feature vectors which in turn used for feature selection in the dimensionality reduction. This method also uses a classification algorithm namely, AdaBoost with random forests (ADBRF) algorithm which builds a classifier that categorize the input image into tumor affected or unaffected. Experimental results of the proposed algorithm are compared to the existing methods on brain MRI segmentation and brain tumor detection shows our method outclasses the other advanced methods.