Diffusion Models in Oncology: a Review of Current Applications

Purpose of Review: Diffusion models are an emerging class of generative artificial intelligence with increasing relevance in oncology. This review examines how diffusion-based methods are being applied across cancer imaging, radiotherapy, and therapeutic development, and discusses their translational potential and key barriers to clinical adoption. Recent Findings: Recent studies demonstrate that diffusion models can generate high-quality synthetic medical images across multiple modalities, improving diagnostic model performance, particularly for rare tumor subtypes. Diffusion-based approaches have shown superior performance in tumor and tissue segmentation, radiotherapy dose prediction, and synthetic CT generation from cone-beam CT for adaptive treatment planning. Beyond imaging, diffusion models have enabled the design of novel small molecules and antibodies, such as those targeting the HER2 antigen, with experimentally validated anticancer activity. However, challenges remain, including high computational demands, inference latency, privacy risks related to data memorization, and limited validation in real-world clinical workflows. Summary: Diffusion models are emerging as a versatile generative framework with measurable impact across multiple domains of oncology. Continued progress in efficiency, validation, and integration will be essential for translating these methods from research settings into routine clinical care.