A Framework for Automatic Detection of Brain Tumor Using Texture Pattern Matrix and Clustering Algorithm

Abstract

ABSTRACT newlineBrain Tumor is a complex disease that occurs due the abnormal growth of brain cells. For newlineefficient treatment planning, earlier detection of tumor is necessary. Magnetic Resonance newlineImaging (MRI) is now recognized as an important tool for the detection of Brain tumor. newlineMRI can be used to identify various tissues inside the brain with good efficiency and accuracy. newlineThe radiation used in MRI is non-ionizing and the contrast agents used are less newlineharmful. Computer Aided Diagnosis (CAD) could be almost as effective as double reading newlineby providing a second opinion to the radiologist, and help in increasing the sensitivity and newlineaccuracy of detection. The proposed Brain Tumor detection algorithm is composed of four newlinestages: Preprocessing, Segmentation, Feature Extraction and Classification. Major steps in newlinepreprocessing are noise filtering, contrast enhancement and skull stripping. A novel algorithm newlinefor brain MRI segmentation using K-Means Clustering and Texture Pattern Matrix is newlineproposed in this work. newlineMedian Filter is used to remove noise and high frequency components from MRI without newlineaffecting its edges and bandwidth. Contrast Limited Adaptive Histogram Equalization newline(CLAHE) is used to enhance brain MRI. Skull stripping is based on connected regions and newlinemorphological operations. K-Means clustering with Texture Pattern Matrix (TPM) based newlinesegmentation process is implemented to detect Brain Tumor. Here, the performance of MRI newlinesegmentation algorithm in terms of accuracy, specificity and sensitivity are computed using newlinemanually segmented ground truth images. In this study Region growing,Watershed and Active newlineContour Model (ACM) were implemented to authenticate the performance of proposed newlinemethod. Fuzzy C-means (FCM) algorithm is also implemented and it is combined with newlineTPM to evaluate the performance of segmentation algorithm. The parameters used to evaluate newlinethe performance of segmentation are Mean Square Error (MSE), Peak Signal to Noise newlineRatio (PSNR), Accuracy, Correlation, Dice Coefficient and Jaccard Index. Gray Level Co- newlineOc