Using a patient-specific diffusion model to generate CBCT-based synthetic CTs for CBCT-guided adaptive radiotherapy

Area of Science:

• Medical Imaging
• Radiotherapy Physics
• Artificial Intelligence in Medicine

Background:

• Cone beam computed tomography (CBCT) is vital for tracking anatomical changes during image-guided radiation therapy (IGRT).
• CBCT artifacts hinder the full potential of adaptive radiation therapy (ART).

Purpose of the Study:

• To improve CBCT image quality for accurate assessment of patient- and fraction-specific (PFS) anatomical variations during ART.
• Enhance the efficacy of radiation therapy (RT) by acquiring high-quality CBCT images.

Main Methods:

• Proposed patient- and fraction-specific lung diffusion models (PFS-LDMs) using a pre-trained general lung diffusion model (GLDM).
• Fine-tuned PFS models on CBCT-deformed planning CT (dpCT) pairs for each patient to learn personalized anatomical changes.
• Evaluated PFS-LDMs using Mean Absolute Error (MAE), Peak Signal-to-Noise Ratio (PSNR), Normalized Cross-Correlation (NCC), and Structural Similarity Index Measure (SSIM).
• Compared PFS-LDMs with a Generative Adversarial Network (GAN)-based model, demonstrating the applicability of the PFS fine-tuning strategy.

Main Results:

• PFS-LDMs significantly improved all four evaluation metrics compared to the GLDM.
• Fine-tuning reduced MAE from 103.95 to 15.96 HU and increased mean PSNR, NCC, and SSIM.
• The PFS fine-tuning strategy enhanced a Cycle GAN model, with PFS-CG models outperforming the general model.
• PFS-LDMs demonstrated superior performance over the GAN-based model across all metrics.

Conclusions:

• PFS-LDMs substantially enhance CBCT image quality, improving Hounsfield unit (HU) accuracy and reducing artifacts.
• Improved image quality facilitates better capture of inter-fraction anatomical changes for CBCT-guided ART.
• This work enables more efficient and personalized high-precision radiation therapy by addressing anatomical variations.