Privacy Preserving with M-SVM Classifier of Tumor Classification in Mammography Images Using Multiple Otsu'S Thresholding Technique

Abstract

Breast Cancer is formed by an abnormal development of cells in breast. The cells of body separate in an incessant method and occupy to surrounding tissues. It is the important reason of death amongst women and after lung cancer breast cancer is second cause of women deaths. Early breast cancer detection can lead to death rate decrease. The mammography is executed to discover the breast cancer tumor at earlier stages. Early breast cancer tumors detection based on the both the radiologists capability to read mammogram images and image quality. The tumors classification is a medical application that set a huge issue for in the breast cancer recognition area. Therefore, in this paper, a multiple otsu's thresholding method is presented with Mutlti-class SVM (M-SVM) classifier to enhance the tumor classification in mammogram images for cancer tumor detection. In this process, elimination of artifacts, noise and surplus parts that are presented in mammogram images by employing preprocessing tasks and after that it improves the mammogram image contrast utilizing CLAHE (Contrast Limited Adaptive Histrogram Equalization) technique for simpler recognition of tumors in breast. We segment the images using Multiple Otsu's thresholding technique to identify the region of interest in mammogram image after preprocessing and image enhancement. The GLDM (Gray Level Difference Method) is exploited to extract the features from the mammogram image. Feature extraction has been employed to with hindsight examine screening mammograms in use prior to the malignant mass discovery for early breast cancer tumor detection. The extracted features can be given to the M-SVM Classifier to classify the tumor in mammogram image into malignant, benign or normal based on the features. The classification accurateness based on the stage of feature extraction. Results of mammogram image is planned by classification and lastly image categorized into Normal, malignant or Benign. Experimental results of proposed method can show that this presented technique executes well with the accurateness of classification reaching almost 84% in evaluation with existing algorithms.