A CAD system for Histological grading using Histopathological Image of Breast Cancer

• Main Authors: Chi-Yang Chen, 陳麒揚
• Other Authors: Chien-Chuan Ko, Ph.D
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/60644667652926727539

碩士 === 國立嘉義大學 === 資訊工程學系研究所 === 104 === Breast cancer is one of the most common disease for women in Taiwan. It also is the second leading causes of death for female. Among the, invasive ductal carcinoma accounts for seventy-two percent of the breast cancer, and causes a big threat to the domestic women. For clinical pathologists, Nottingham Modification of the Bloom-Richardson System is a gold standard based on histological pathology to evaluate the invasive ductal carcinoma. The grading indices used include tubular formation, plemorphism, and mitotic count. The major goal of this research is to develop a computer-aided-diagnosis system to assess the severity of breast carcinoma. It analyzes the H&E stained slide images of breast specimen to extract feature parameters related to morphometry of mammary duct, and hyperplasia degrees of nucleus, and mitotic count of nuclei based on histology and cytology. Moreover, the proposed system provides histological grade and prognosis for clinical pathologists to improve the efficiency of diagnosis. In this study, high and low magnification histology-slide images of breast tissue specimens fetched from a microscope were selected carefully for our experimental samples. For low power images, a series of image processing operations including color transform, smoothing, K-means clustering, and watershed operation were used to separate the mammary ducts of interest. Next, the shape and texture features combined with a classifier were extracted to evaluate the severity of tubular formation. For high power images, the nuclei of interest related to the analyses of pemprphism, and mitosis count of nuclei were segmented using Expectation-maximization algorithm and ellipse fitting. Next, feature parameters including area and gray level difference were measured for the following classification stage. Finally, the feature parameters extracted from abnormal cell and the mammary duct were trained and validated based on support vector machine (SVM) after feature selection stage in order to choose a better classification model based on receiver operating characteristic curve analysis. Experimental results demonstrate that the identification accuracies of the histological pathology were 90.12% in tubular formation, 88.75% in nuclear pleomorphism, 88.81% in mitosis count, respectively. The developed grading system not only can be used to evaluate the severity of the breast carcinoma, but also can provide references for the prognosis analyses or treatment of the breast carcinoma.