Construction of Classifier of Tumor Cell Types of Pancreas Cancer Based on Pathological Images Using Deep Learning

Abstract

Recently, computer-aided diagnosis methods based on machine learning, mainly using Deep Leaning, have been studied and developed very rapidly. Especially image recognition based on Convolutional Neural Networks showed high accuracy in diagnosis problems when they are given the huge amount of training data. Those methods are not only helpful for classification but also useful for feature extraction from given images. Here we introduce a new classification method to find the features of tumor tissues from histopathology images by unsupervised clustering based on Information Maximization Self-Augmented Training. Moreover, to evaluate fibrosis and classify tumor cells, we used histopathological images with different staining methods as concatenated inputs. Using this approach, we can quantify integrated features based on multimodal imaging using deep learning. In this study, we analyzed pathological images of pancreas cancers and optimized to classify the patches of the images into the categorize with different features, which are consistent with annotation of the medical doctors. It can also provide a map to visualize the probability where cell types are categorized into specific classes according to the given pathological images.