Conditional Diffusion Models for CT Image Synthesis from CBCT: A Systematic Review

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Cone Beam Computed Tomography (CBCT) offers low-dose volumetric imaging for radiotherapy but suffers from image quality issues like noise and artifacts, limiting its use for synthetic CT (sCT) generation.
• Conditional Diffusion Models (CDMs) are emerging deep learning techniques that may overcome these limitations due to their iterative denoising capabilities, potentially preserving anatomical details and modeling uncertainty better than previous methods.

Purpose of the Study:

• This systematic review critically evaluates the application of CDMs for CBCT-to-CT synthesis.
• It focuses on identifying architectural strategies, analyzing reported quantitative outcomes, and assessing the clinical relevance of these models.
• The review addresses the types of CDMs used, their performance metrics, and their discussed clinical implications.

Main Methods:

• A comprehensive systematic literature search was performed across major scientific databases (PubMed, Web of Science, Scopus, IEEE Xplore, Google Scholar).
• Studies published between 2013 and 2024 were screened for eligibility.
• Eleven studies meeting the inclusion criteria were analyzed to answer the review's core questions.

Main Results:

• Conditional Diffusion Models demonstrated promising image quality in CBCT-to-CT synthesis, particularly when enhanced with anatomical priors, spatial-frequency guidance, hierarchical refinement, or latent representations.
• However, the current evidence base is limited by heterogeneity in anatomy, dimensionality, supervision, and evaluation metrics, restricting definitive comparative claims.
• The reviewed studies suggest CDMs are a promising avenue for sCT generation, but require further dose-aware validation, standardized reporting, and multicenter evaluation.

Conclusions:

• Conditional Diffusion Models represent a significant advancement in the field of CBCT-to-CT synthesis, offering improved image quality over traditional methods.
• Despite promising results, the heterogeneity and limitations of current studies necessitate further research before widespread clinical adoption.
• Future work should focus on robust validation, standardization of evaluation, and multicenter trials to establish the clinical utility of CDMs for synthetic CT generation.