ResNet-50 based Method for Cholangiocarcinoma Identification from Microscopic Hyperspectral Pathology Images

As the second most common primary liver tumour, the early detection of cholangiocarcinoma is very important. Computer-aided diagnosis based on deep learning using pathological tissue images is often used in cancer diagnosis. Compared with traditional RGB pathological images, hyperspectral image has more advantages in deep learning based automatic pathological diagnosis because it contains spectral dimension information. In this paper, a ResNet-50 based method is used to identify cholangiocarcinoma from microscopy hyperspectral images. The microscope hyperspectral choledoch tissue images are captured by our microscopy hyperspectral imaging system (MHIS) and annotated by experienced pathologists manually. After pre-processing and data argumentation, we split them in to training set (6800 images) and testing set (210 images) and choose ResNet-50 structure to train the classification model. The classification model can automatically classify the choledich tissue images into cancerous and non-cancerous regions. Our experimental results show that the accuracy of proposed method is 82.4% in case of ResNet-50 structure.