3D Latent Diffusion Model for MR-Only Radiotherapy: Accurate and Consistent Synthetic CT Generation.

Area of Science:

• Medical Imaging
• Radiotherapy Planning
• Artificial Intelligence

Background:

• Clinical need to reduce ionizing radiation exposure drives demand for synthetic imaging solutions.
• Existing cross-modal synthesis methods (GANs, CNNs) struggle with high-contrast tissues, limiting reliability.
• Safety-critical applications like radiotherapy require high-fidelity synthetic CT (sCT) generation.

Purpose of the Study:

• Develop a stable, high-accuracy framework for 3D MRI to CT synthesis.
• Generate clinically equivalent sCTs across diverse anatomical regions.
• Establish a robust solution for radiation-free medical imaging.

Main Methods:

• Introduction of a novel 3D Latent Diffusion Model (3DLDM).
• Operation within a compressed latent space to manage computational load.
• Leveraging a stable denoising objective for improved synthesis.

Main Results:

• 3DLDM achieved a Mean Absolute Error (MAE) of 56.44 HU across Head & Neck, Thorax, and Abdomen.
• Demonstrated significant error reduction compared to CycleGAN (3.63%), NNUNet (10.76%), and SwinUNeTr (20.79%).
• Achieved superior structural similarity (SSIM) and peak signal-to-noise ratio (PSNR) over existing methods.

Conclusions:

• Validated a scalable and accurate approach for volumetric image synthesis.
• 3DLDM enables MR-only radiotherapy planning, reducing patient radiation dose.
• Accelerates clinical adoption of radiation-free multi-modal imaging.