Classification of Brain MRI Images Based on Segmenting Tumor ROIUsing Histogram Based Multi-Component &Multilevel Adaptive Thresholding

Abstract

The work presented in this paper is focused on extracting the tumor region from Brain MRI images and then classifying them as benign or malignant. The first part comprises detection of tumor part using histogram based multi-component and multilevel adaptive thresholding technique and the later includes tumor and non-tumor part features to classify the MRI images in two classes. The multi-component approach uses all the three components other than the approximation component of wavelet decomposed image for thresholding and the multi-level approach includes two level wavelet decomposition of the original gray level MRI image.The 6-two-dimensional component values at both the levels are transformed and histogram corresponding to each component are down sampled to level 5 using 1D wavelet transform. The Global minimal value at level 5 is the suitably used as a threshold value to filter each individual component and the new image is then reconstructed. Finally, a mean threshold is used to segment the tumor part from the MRI image. The classification using three different classifiers is based on features extracted from the tumor and non-tumor part. Out of 77 normal and 143 affected images, the proposed technique was able to classify the images with greater accuracy.