Automatic recognition of tumor region in multiphoton images of hepatocellular carcinoma using a convolutional neural network

The fundamental principle of hepatectomy is to entirely excise the tumor while preserving adequate functional liver tissue volume. Thus, identifying tumor and non-tumor areas swiftly can enhance the precision and efficiency of liver resection, ultimately improving patient survival rates. In this study, we utilized multiphoton microscopy (MPM) to label-free identify liver tumor and non-tumor regions, following by automated classification with an open-source convolutional neural network, ResNet. The outcomes demonstrate that the network model can automatically and effectively distinguish tumor and non-tumor regions without human recognition, and MPM combining with deep learning may serve as an auxiliary tool for rapidly detection of hepatocellular carcinoma and aiding in liver resection treatment.