Research On Tumor Regions Segmentation Of Breast Digitized Mammograms Based On Multilevel Otsu Thresholding

Breast cancer is the most cancer for women.Mammography is a very reliable and practical method for breast cancer detection and analysis at an early stage.Locating and segmenting tumor regions in digitized mammograms are key steps for breast cancer diagnosis.In this paper,we present a tumor segmentation method based on multilevel Otsu thresholding by analyzing relative references and comparing with the state-of-the-art methods.This could rapidly and effectively segment tumor regions and has a great potential to assist physicians for diagnosing breast disease.Digitized mammograms segmentation includes three steps: pre-processing,tumor regions segmentation,post-processing.In the pre-processing step,we adopt the mean filtering and the traditional Otsu to reduce image noise,remove image label and locate breast regions.In the tumor regions segmentation step,we present a multilevel Otsu thresholding to determine final tumor segmentation regions.In the post-processing step,the morphological algorithm is used for the smoothing contours of the tumor regions.Hereinto,the multilevel Otsu thresholding is the most significant algorithm which overcomes the limitations of the traditional Otsu,satisfies the segmentation demands of the complex images and improves the segmentation accuracy rates of the digitized mammograms.To verify the effectiveness of the method,we choose 74 images of the MIAS database from different kinds of tumors and four prevalent metrics including Variation(VAR),Uniformity Measurement(UM),Contrast Measurement(CM),Overlap(OV).What is more,Maximum entropy,2-mode,iteration and mathematical expectation are determined as comparing methods in the experiments.Our method then has a competitive performance compared with the state-of-the-art algorithms.