Deep learning for liver cancer histopathology image analysis: A comprehensive survey

Liver cancer is the predominant cause of cancer-related fatalities globally, wherein Hepatocellular Carcinoma (HCC) and Intrahepatic Cholangiocarcinoma (ICC) emerge as the principal subtypes. Histopathology images, revered as the definitive benchmark for liver cancer diagnosis, yield rich phenotypic information, instrumental in facilitating disease progression prediction and potential survival prognostication. Deep learning has been rapidly developed recently and has become the mainstream technique for liver cancer histopathology image analysis, showing noteworthy accomplishments. This article undertakes a comprehensive examination of over 50 publications within the domain of deep learning-based liver cancer histopathology analysis, systematically discussing many advanced approaches. We commence our exploration by elucidating diverse facets of this field, encompassing problem formulation, general learning paradigms, and main challenges. Subsequently, we present a meticulous summary of publicly accessible datasets and evaluation metrics. To foster a deeper understanding of the research status of this domain, we furnish a taxonomy covering supervised learning and weakly supervised learning approaches within the specific tasks, i.e., classification and localization for histopathology diagnosis as well as deep learning-based survival models for disease prognosis. Finally, we discuss existing open issues and potential future trends within the realm of computational histopathology in liver cancer research.