Tumor Demarcation by using Local Thresholding on Selected Parameters obtained from Co-occurrence Matrix of Ultrasound Image of Breast

Ultrasound imaging (US) is the most widely used and important imaging modality in medical domain. Due to certain artifact such as speckle, segmentation of US image has not remained a trivial task. Two stages segmentation process has been used in this paper to detect the solid mass (cancer) in breast US image. GLCM based texture feature image generation followed by local adaptive thresholding. In first, Correlation, Variance, Sum variance and Sum average texture features for all angular relationships has been implemented on original image to obtain the feature images. In Second, adaptive local thresholdinig algorithm is applied recursively by dividing the feature image into nine sub-images and compared with the result of Otsu's global thresholding technique. Results of our algorithm are better.