A Method Of Weak-Supervised Morphology Classification For Imprint Cytology Of Breast Cancer

During breast cancer surgery, intraoperative diagnosis is required to determine surgical margins. However, the current methods for determining margins are time-consuming. A new diagnostic method using ”click-to-sense” probe allow the rapid, selective and sensitive diagnosis. In addition, the automation of diagnosis will greatly assist in intraoperative decision making. Imprint cytology images are large as whole slide histopathology images and manual annotation is time-consuming due to the size and requirement of expertise for cytology. Only sample-level labels can be obtained from careful diagnosis. Because no a-priori knowledge of which patches within them are associated with the label, this situation is known as weakly supervised learning. In this study, we propose a classification method consists of DeepCluster and k-nearest neighbor (kNN) classifier.