Brain Tumor Diagnosis Using MR Image Processing

Abstract

Magnetic Resonance Imaging (MRI) images of brain are a process of high importance to diagnosing brain tumors. Brain tumor is an abnormal growth of cells in the brain. These tumors may be benign or malignant. The use of computer technologies became widely used to store and manage medical images for supporting medical decision and improve the accuracy of radiologists with a reduction of time in the interpretation of images. The present study aimed to establish a Computer-Aided Detection and Diagnosis (CADD) system dealing with medical MRI for classifying input digital image into normal or abnormal tumors, also the type of abnormal case is diagnosed into benign or malignant tumor. The proposed method is considered to be contained four stages within, they are: Pre-processing stage, image segmentation for determining the Region of Interest (ROI), Feature extraction based on Scale Invariant Feature Transform (SIFT) descriptor, and then classification. The results of classification are evaluated by cross validation technique, in which the dataset are divided into training set and testing set. To evaluate the achieved results, the classification is carried out using two levels for each case: logistics technique is used to check the results of normal case, and random forest to check the results of abnormal cases. Results of normal classification showed that the accuracy of applying Logistic Regression was 93.3%, whereas the classification score of abnormal cases was 99.9% for Random Forest, which ensure the success of the classification system and correct path of the computations.