Font Size: a A A

Breast Anatomical Markers Detection And Its Application In Region Matching

Posted on:2020-10-18Degree:MasterType:Thesis
Country:ChinaCandidate:R M QinFull Text:PDF
GTID:2404330590458401Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Radiologists often compare the corresponding regions in different views of mammograms when they utilize mammogram to examine breast cancer.Meanwhile,the anatomical landmarks in breast are the most important references for multi-view region matching.However,the pressed and deformed during the breast shooting presents a great challenge to detecting breast anatomical markers accurately.The anatomical landmarks in the breast mainly include breast boundary,pectoral muscle edge line and nipple.The breast boundary is extracted by threshold segmentation and boundary tracking.Then different methods are adopted to remove the boundary of nonbreast tissues(arms,armpits,artifacts,muscles)according to the boundary feature,such as the boundary morphological characteristics and the gray distribution around the breast boundary.Moreover,the pectoral muscle edge line detection uses a method based on evaluation strategy,which solves the problems of detecting pectoral muscle edge lines from those pectoral muscles which are stratified,have inconsistent texture or have gray value very close to the dense gland tissue.Specifically,the proposed method segments the pectoral muscle tentatively by different thresholds after reducing the gray level via blocking,then extracts and selects the pectoral muscle edge pixels to fit the pectoral muscle edge line as a straight line.The same procedure is performed with other thresholds to obtain multiple candidates of pectoral muscle edge lines,finally the optimal one is determined from the candidates according to pectoral muscles morphological characteristics.To detect nipple in different situation,an algorithm is proposed to ensure that the most significant features are used preferentially.For the prominent nipples,the concave-convex feature of breast boundary is used to detect.And for the inverted nipples,the proposed algorithm detects them with the characteristics of brightness.Furthermore,for images without visible nipple,the nipples are located as the farthest point from pectoral muscle edge line.After extracting the three anatomical markers for breast,a normalized coordinate system which refers to the anatomical markers is established to complete multi-view region matching.The method is validated with 768 mammograms provided by United Imaging Healthcare.Experimental results show that acceptable rate of the pectoral muscle edge line is 97.5%.At the same time,acceptable rate of the nipple is 97.7%,and the error of nipple detection is about 1.8±2.65 mm.
Keywords/Search Tags:Mammogram, Breast boundary, Pectoral muscle, Nipple, Multi-view region matching
PDF Full Text Request
Related items