The field of lung cancer diagnosis and risk prediction is rapidly advancing with the development of innovative deep learning models. Recent research has focused on improving the accuracy and efficiency of lung cancer diagnosis, particularly in the early stages. One of the key trends is the use of generative models and hybrid architectures that combine different techniques, such as autoencoders, transformers, and attention mechanisms, to analyze medical images and predict cancer risk. These models have shown promising results in detecting malignant nodules, classifying tumor stages, and predicting lung cancer risk from low-dose CT scans. Notably, some studies have demonstrated the potential of these models to match or even surpass the performance of traditional clinical follow-ups and human experts. The use of anatomy-aware and clinically informed approaches has also led to more transparent and interpretable decision support systems. Overall, the field is moving towards more accurate, efficient, and clinically relevant deep learning models for lung cancer diagnosis and risk prediction. Noteworthy papers include: LungX, which achieves state-of-the-art performance in pneumonia detection with a hybrid EfficientNet-Vision Transformer architecture. LungEvaty, a scalable and open-source transformer-based model for lung cancer risk prediction that operates on whole-lung inputs and matches state-of-the-art performance.