Prediction of breast cancer metastasis by deep learning pathology
Yuanyue Lu, Jun Zhang, Xueyu Liu, Zhihong Zhang, Wangxing Li, Xiaoshuang Zhou, Rongshan Li
Abstract
Abstract With the rapid development of social economy, the incidence of breast cancer is increasing year by year. Whether there is lymph node metastasis in frozen tissue sections during breast cancer surgery is of tremendous priority for breast cancer surgical decision‐making. Therefore, it is very significant to diagnose the pathological sections of breast cancer quickly and accurately. In this study, a model which can quickly fine segmentation of lesion regions in high‐resolution breast cancer pathology sections is proposed. Firstly, pathology sections are processed by pre‐processing module; Secondly, the main lesion region in pathology sections can be quickly recognized by recognition module; Thirdly, the fine segmentation of lesion region can be accomplished by segmentation module. The dataset is selected from two medical institutions to evaluate the proposed model; it achieved the average recognition precision of 0.936 for region of interest in high‐resolution pathology section, with an F1‐score of 0.787; and the dice for lesion region segmentation is 0.8517. The proposed model outperforms several similar works, which can effectively improve the speed and precision of pathologist's diagnosis for high‐resolution breast cancer pathology sections.