Computer-Aided Detection System of MRI brain tumor images

Abstract

Computer-Aided Detection (CAD) recognizes tumors or lesions in medical imaging or distinguishes between normal and abnormal images. The purpose of this paper is designing a CAD system that will automatically detect brain tumors and classify the brain images in terms of normality and abnormality. The proposed CAD system passed through seven essential processes which are data collection, preprocessing and enhancement, segmentation, feature extraction, feature selection, classification, and performance assessment, respectively. The database includes 280 normal and abnormal brain MRI images. Segmentation process in this paper was an independent process aims to aid in the extraction of the region of interest (ROI). ROIs were cropped from the original images around the center of the tumor region which was specified after segmentation. The overall results of the proposed CAD system depended on the performance of eight different types of SVM classifiers and KNN classifiers. SVM of radial basis function and linear types, as well as KNN of 3 and 5 neighbors, obtained perfect results with 100% in all performance assessment metrics. The remainder of the classifiers achieved high accuracy, where SVM of polynomial type with KNN of 1 and 2 neighbors achieved the same result with 97.62% a little less than KNN of 4 neighbors which achieved 98.81%. The proposed CAD system provided results more accurate and precise compared with other studies.