Synthetic CT image generation for CBCT-guided lung radiotherapy using a diffusion model

Le résumé fourni par la source

Despite having lower image quality than planning CTs (pCT), cone beam computed tomography (CBCT) is widely used in the clinic to guide the radiation delivery due to its efficient scanning time. The so-called, image-guided radiation therapy (IGRT) leverages CBCT to minimize patient positional uncertainty for modern radiotherapy. However, CBCT image quality can be compromised by severe stripe and scattering artifacts, as well as by tumor movement due to respiratory motion. To conquer those challenges, we propose a lung diffusion model (L-DM) framework to improve CBCT image quality. This model uses a conditional diffusion model trained on pCT and deformed CBCT (dCBCT) image pairs to synthesize lung CT images from dCBCT images, improving CBCT-based IGRT. We validated our model using an institutional lung cancer dataset, which demonstrated significant improvement. Our L-DM method reduced the mean absolute error (MAE) from 99.44 Hounsfield units (HU) to 36.69 HU, increased the peak signal-to-noise ratio (PSNR) from 22.85 to 29.16 dB, improved the normalized cross-correlation (NCC) from 0.83 to 0.98, and enhanced the structural similarity index measure (SSIM) from 0.82 to 0.93.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.