Feature Extraction and Feature Selection Methods in Classification of Brain MRI Images: A Review

Abstract

Magnetic Resonance Imaging, or MRI, is one of the safest ways to observe human tissues and organs, like the brain. It causes no radiation and bad effect for the patient. The necessity of early detection of brain tumor leads the researchers to initiate this study. Early detection of a brain tumor can be observed through MRI image results. The brain MRI images are then processed by using 3 major steps, namely image processing, feature extraction and selection, and classification. The feature extraction and selection are one of the important steps that can determine the accuracy of brain MRI images classification in which its result will determine the disease. This paper examines 9 feature extraction and 3 feature selection methods for classification of brain MRI images. Furthermore, it explores the advantages and disadvantages of each method. Both of them are considered in the selection of the best method to be applied in different cases. The summary of each method is presented in a table as supportive evidence. This study shows that Local Binary Pattern combined with GLRL, ZM, PHOG, and GLCM is the best feature extraction method for BRATS dataset with a classification accuracy of 97.1%, while GLDM and GA are the best combinations of feature extraction and selection method for clinical datasets with a classification accuracy of 98%.