Brain tumor pixels detection using adaptive wavelet based histogram thresholding and fine windowing

In recent years, image processing had covered a wide area over medical applications in diagnosis of wide variety of diseases in medical images. Brain tumor detection is one of the most widely used applications by vast researchers. In this paper a novel approach is presented for detection of affected mass (tumor) in magnetic resonance images (MRI). Two level wavelet transform is used to decompose the brain image with mother wavelet `db6'. The horizontal, vertical and the diagonal components at both the levels are further decomposed to level five using histogram for each of the component using one dimensional wavelet transform. By finding global minima at level five, and then mapped at the component histogram, a threshold value is found out. The component then is thresholded using this adaptive threshold. The original image is then reconstructed with the obtained components using this course segmentation. The approximation component at level one is unaltered. Lastly a windowing technique is used to eliminate the false detection due to light in-homogeneity or due to hard tissues, which uses a window based threshold. Results after fine segmentation showed that normal patient images showed complete black region in the final segmented image whereas malignant images showed the region of interest after segmentation. The MRI images from a research institute were obtained and the results after segmentation were validated from an expert from the same research institute.