ABSTRACT: Hepatocellular cancer (HCC) is the fifth most common malignancy and the fourth leading cause of cancer-related deaths worldwide. Liver histology plays a crucial role in the biological profiling of HCC, informs cancer prognosis, and supports personalized treatment strategies. Over the past decade, machine learning, deep learning algorithms, and convolutional neural networks have emerged as powerful tools for the histological and molecular assessment of malignancies. Machine learning and deep learning algorithms applied to whole slide digital images (WSIs) of liver cancers have demonstrated significant accuracy in distinguishing non-cancer versus cancer liver tissue and histologically subtyping HCC phenotypes associated with different clinical outcomes. Generative Artificial Intelligence models applied to WSIs obtained from hematoxylin and eosin-stained (H&E) histology specimens have shown promising results in delivering crucial insights into the genetic HCC disarrangement, potentially providing the biological rationale for molecular-targeted therapeutic strategies. This review highlights the diagnostic advances in computational histology for primary liver cancer. The manuscript focuses on the state of the art in AI-based histotyping and molecular profiling for HCC. A critical evaluation of their current performance is essential for inspiring the clinical research priorities and promoting the safe employment of AI models in managing HCC patients.

Impact statement: Histology-based AI applied to H&E wholeslide images improves HCC diagnosis, phenotyping, molecular inference and prognostication, strengthening risk stratification and personalized care, and accelerating integration of computational pathology into routine clinical workflows.

Key words: Artificial intelligence; liver cancer; hepatocellular carcinoma; deep learning; machine learning.