Brain Tumor Classification and Watermarking of MRI Using Nonsubsampled Contourlet Transform

Abstract

Automatic or semi-automatic brain tumor classification scheme is demanded in today’s medical system to get rid of human involvement of classification of brain tumor images. So, here we propose nonsubsampled contourlet transform (NSCT) based MRI brain tumor classification using support vector machine (SVM) and artificial neural network (ANN) classifier. In this scheme, K-means clustering is used for segmentation of region of interest. NSCT is applied to the region of interest of brain image in order to obtain its low and high subband coefficients. Then from the coefficients of NSCT, twelve features are extracted from the region of interest. SVM which incorporates two stages is trained with these twelve features. 1st stage of SVM is able to classify brain image as normal or abnormal and then 2nd stage of SVM classifies grade of tumor as low grade, where tumor is slowly growing or high grade, where tumor is rapidly growing. The grade of tumor is also classified using the ANN classifier based on feed forward back propagation. Furthermore, when the multimedia contents like MRIs or other images are transferred through a communication channel, sometimes the whole content or its part may be modified or deteriorated by hackers. In order to protect this content from unauthorized user, digital watermarking is considered to be a promising tool. So another purpose of this research is to develop image watermarking scheme which ensures higher security, imperceptibility and robustness against different distortion attacks. In first proposed scheme of image watermarking, NSCT is also used because most of the perceptual content of an image focuses on low frequency subband of NSCT. Singular value decomposition (SVD) is also applied on low frequency subband of NSCT, because the singular values taken from low frequency subband have certain stability. Besides, game of life (GOL) cellular automata is used to scramble binary watermark so that no one can recover the watermark without secret scrambling keys. So in this scheme, NSCT and SVD ensure the imperceptibility and robustness as well as cellular automata improves the security. In second proposed scheme of watermarking, multiple chaotic maps, NSCT and discrete cosine transform (DCT) are used. Here, an arranged chaotic sequence which is created by logistic map is used to shuffle the pixel positions of MRI. Patient information, watermark is encrypted by two chaotic maps, like Arnold’s Cat map and tent map. Then, DCT coefficients of encrypted watermark are embedded into the DCT coefficients of NSCT’s approximation band of shuffled MRI. Both proposed watermarking schemes are tested on varieties of MRIs and their associated results reveal that the two schemes have promising improvements in imperceptibility and robustness against noise and geometric attacks. Overall, NSCT is used in both brain tumor classification and watermarking in this research.