Automatic detection and categorization of brain tumor types and grades from MRI images

Abstract

newline Brain tumor, otherwise known as intra cranial neoplasm is the anomalous growth of newlinecells within the brain structure. In recent years, medical images have found newlinecomprehensive applications in healthcare for diagnosing various diseases, therapy newlineplanning and monitoring the disease progression; that incorporates image newlineacquisition of the troubled organ using diverse modalities. When compared to the newlineother image modalities, magnetic resonance imaging (MRI) is an essential tool in newlinethe clinical as well as operational scenario on account of its ability for soft tissue newlinediscrimination, good spatial resolution and better contrast. Moreover, it does not newlinehave any radiation problems. One of the most recent research directions is to further newlinecategorize the brain tumor type into four different grades. Recent advances in newlinemedical image processing with the application of computer aided diagnosis (CAD) newlineusing MRI facilitates easy detection of the tumor portion. Nevertheless, newlineidentification and classification of brain tumor types and grades is still a challenging newlineproblem. Hence, an efficient automated medical decision support system is quite newlineinevitable for detection, classification and analysis of MRI brain images. newlineIn the first phase of the study, a feature reduction technique is proposed and newlineautomatic classification of brain tumors using conventional supervised classifiers is newlineperformed. Here, MRI brain images are classified into normal and abnormal newlinecategories. The feature extraction is done with gray level co-occurrence matrix newline(GLCM) and then feature reduction is performed based on statistical test which is newlinepreceded by principal component analysis (PCA). Main focus of the work is to newlineestablish statistical significance of the features obtained after PCA and a newlinecomparative study using k-nearest neighbour (k-NN), support vector machine newline(SVM) and artificial neural network (ANN) based supervised classifiers is then performed.