AUTOMATED CLASSIFICATION FOR BRAIN MRIS BASED ON 2D MF-DFA METHOD

Abstract

Magnetic resonance image (MRI) is an important tool to diagnose human diseases effectively. It is very important for research and clinical application to classify the normal and abnormal human brain MRI images automatically. In this paper, an accurate and efficient technique is proposed to extract features of MRIs and classify these images into normal and abnormal categories. We use two-dimensional multifractal detrended fluctuation analysis (2D MF-DFA) to obtain image features. These features are the local generalized Hurst exponents calculated by 2D MF-DFA. In this regard, the values of Hurst exponents are given as the training input vector and are taken to the classifiers. We use k-nearest neighbor (k-NN) and support vector machine (SVM) to classify a specific brain MRI as normal or glioma affected. For SVM, we apply the leave-one-out cross-validation method for experimental verification. The 2D MF-DFA-SVM system achieved accuracy, sensitivity, and specificity of 99.82% +/- 0.07, 100%, and 99.81% +/- 0.09, respectively. The 2D MF-DFA-k-NN system achieved accuracy, sensitivity, and specificity of 96%, 92.59%, and 100%, respectively. We find that when performing binary classification for brain MRIs, the SVM is superior to k-NN. In addition, our experimental results indicate that the proposed 2D MF-DFA-SVM achieved excellent outcomes compared to those of the previous works. The proposed system is a promising system to clinical use.