Abstract
The deep neural network is a research hotspot for histopathological imageanalysis, which can improve the efficiency and accuracy of diagnosis forpathologists or be used for disease screening. The whole slide pathologicalimage can reach one gigapixel and contains abundant tissue feature information,which needs to be divided into a lot of patches in the training and inferencestages. This will lead to a long convergence time and large memory consumption.Furthermore, well-annotated data sets are also in short supply in the field ofdigital pathology. Inspired by the pathologist's clinical diagnosis process, wepropose a weakly supervised deep reinforcement learning framework, which cangreatly reduce the time required for network inference. We use neural networkto construct the search model and decision model of reinforcement learningagent respectively. The search model predicts the next action through the imagefeatures of different magnifications in the current field of view, and thedecision model is used to return the predicted probability of the current fieldof view image. In addition, an expert-guided model is constructed bymulti-instance learning, which not only provides rewards for search model, butalso guides decision model learning by the knowledge distillation method.Experimental results show that our proposed method can achieve fast inferenceand accurate prediction of whole slide images without any pixel-levelannotations.