Abstract: Lung cancer is one of the malignancies with the highest incidence and mortality rates worldwide, and accurate histopathological subtyping is essential for developing individualized treatment strategies. Conventional histopathological examination has several limitations, including its invasive nature, lengthy turnaround time, and susceptibility to subjective interpretation. The integration of deep learning (DL) with medical imaging provides a novel approach to the histopathological subtyping of lung cancer, effectively improving both efficiency and accuracy. The authors summarize recent progress in CT- and PET-based DL for the histopathological subtyping of lung cancer, focusing on the model architectures and modeling strategies employed in recent studies, including technical approaches represented by multimodal fusion and Transformer-based architectures, and further discuss key issues related to model interpretability, generalizability, and clinical deployment, as well as future directions for the development of DL in this field.