EqDiff-CT: Equivariant Diffusion Model Improves Head and Neck CT Synthesis from CBCT

A new diffusion model (EqDiff-CT) synthesized head and neck CT from CBCT with substantial gains in structural fidelity and Hounsfield unit accuracy compared to CycleGAN and DDPM, potentially improving CBCT-guided treatment planning.

• Uses a group equivariant conditional U-Net with steerable layers to enforce rotational equivariance (C4 symmetry), preserving fine details.
• Validated on the SynthRAD2025 dataset, outperforming CycleGAN and DDPM.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.