Multilevel thresholding and fractal analysis based approach for classification of brain MRI images into tumour and non-tumour

Abstract

In this paper, a method is proposed for classification of brain magnetic resonance imaging (MRI) images as tumour and non-tumour. A multilevel thresholding is used for segmentation. Thresholding is applied to convert MRI images to binary images. Fractal texture analysis is carried out for texture feature extraction. Mean and area features are extracted from binary images. We have computed fractal dimension (FD) using box counting method. The fractal measurements describe the boundary complexity of objects and structures beings segmented. Three features extracted, namely, mean, area and FD are used for classification. The images are classified as tumour or non-tumour using artificial neural network (ANN). The experiments are carried out on coronal, sagittal and axial views of brain MRI images. We have used the different number of thresholds (t) in the range [0-10]. We have found that the required value of t is three. Eight different parameters viz. specificity, sensitivity, accuracy, false positive rate (FPR), positive predictive value (PPV), negative predictive value (NPV), false discovery rate (FDR), F-SCORE for optimum number of thresholds are evaluated. We have obtained 100% classification accuracy for all the views of brain MRI images.