An automated diagnosis system using wavelet based SFTA texture features

This paper presents a novel feature extraction technique for an automatic classification of MRI brain tumor images. Wavelet based SFTA Feature extraction technique (WSFTA) consists of two steps: (i) Initially the input image is decomposed into different frequency sub band images using 2-D Discrete Wavelet Transform, (ii) the texture features are extracted from the decomposed Low frequency image using SFTA (Segmentation based Fractal Texture Analysis). Two layered feed forward neural network is used for the classification of MRI brain images into normal or abnormal stages. Performance of this proposed technique is compared with GLCM and Haralick's texture feature in terms of MSE and classification accuracy. It also gives better accuracy and reduced MSE of 98.0% and 0.585 respectively. Thus, the resultant WSFTA technique performs more accurately than the previous works.