Diffusion Models for CBCT-based Synthetic CT Generation in Adaptive Radiotherapy: A Review

Background: Diffusion-based generative models have emerged as a powerful approach for cone-beam computed tomography (CBCT)-to-CT synthesis, addressing a key technical barrier to adaptive radiotherapy. By modelling image generation as an iterative denoising process, diffusion models enable high-fidelity synthesis of CT-like images from on-treatment CBCT while preserving patient- and fraction-specific anatomy. This capability is particularly relevant for adaptive radiotherapy, where up-to-date anatomical information and accurate electron density estimation are essential for reliable dose recalculation and treatment plan adaptation, allowing better patient outcomes. Purpose: This review summarised and critically examined the current state of diffusion models for synthetic CT (sCT) generation, covering methodological developments, conditioning strategies, data requirements, evaluation practices, and clinical relevance. Methods: Using PRISMA 2020 guidelines, a search of PubMed, Embase, and Web of Science identified 13 relevant English-language studies (from an initial 61) on diffusion-based CBCT-to-CT synthesis published before November 2025. Results: Across multiple anatomical sites, diffusion-based approaches consistently demonstrated improved image quality, Hounsfield Unit fidelity, and structural preservation compared with earlier deep learning methods. Several studies further reported clinically acceptable dose recalculation accuracy on diffusion-generated sCT, highlighting their potential to support adaptive treatment planning. However, the review also identified important gaps that limited clinical translation, including reliance on imperfectly aligned training data, limited uncertainty quantification, computational inefficiency for fast synthesis in online adaptive radiotherapy workflows, and a lack of standardized dosimetric and human expert evaluation. Conclusions: By synthesizing recent advances and outstanding challenges, this review provided a comprehensive perspective on the role of diffusion models in CBCT-based adaptive radiotherapy and outlined key directions for future research to enable safe, efficient and reliable clinical deployment.