Automated Diagnosis On Tissue Microarrays Of Breast Cancer

In clinical hospital, there is a Nottingham Histologic Score system available for determining the grade of breast cancer. In this scoring system, there are three factors that the pathologists take into consideration: the amount of gland formation, the nuclear features and the mitotic activity. Due to differences in experience and knowledge, different physicians may have a different score on the histopathological score. The significance of research on the diagnosis of breast cancer based on tissue microarrays is particularly important, because it provides objective and practical criteria for clinicians with different experiences, while avoiding some missed cases.The epithelial tissue and stromal tissue are the two most basic tissues in mammary gland tissue. Studies have shown that about 80% of breast cancers originate from the epithelial tissue, so the epithelial tissue, the stromal tissue and their microenvironme-ntal analysis are important markers for assessing the risk of breast cancer. Therefore,the precise segmentation of epithelial tissue and stromal tissue is a prerequisite for the construction of computer-aided diagnostic system.In view of the above problems, this essay realizes the automatic diagnosis algorithm based on the tissue microarrays. Firstly, we use the fully convolutional networks to segment the small size of epithelial and stromal tissue quickly. The results show that the hightest segmentation accuracy on the VGH and NKI dataset(Pixel accuracy: 92.0%, Mean accuracy: 85.7%, Mean IU: 79.5%, Frequency weighted IU:86.1%), the best segmentation and the segmentation speed (0.097 seconds) under the same conditions. Then through the slide window way, we input lots of blocks to that network to achieve the large-size tissue microarray of segmentation. Finally, the segmentation of the epithelial and stromal tissue are extracted the color features and texture features respectively. And the combination of input features into the support vector machine classifier, then it obtains the pathological grading of results (the Grade I of classification accuracy is 81.7%, the Grade II of classification accuracy is 80.6%). For the actual breast cancer tissue microarray data set to achieve breast cancer histopathological automatic grading, so as to achieve the goal of computer-assisted diagnosis of breast cancer.