Synthetic CT generation from MRI using 3D transformer‐based denoising diffusion model

Abstract

Background and purposeMagnetic resonance imaging (MRI)‐based synthetic computed tomography (sCT) simplifies radiation therapy treatment planning by eliminating the need for CT simulation and error‐prone image registration, ultimately reducing patient radiation dose and setup uncertainty. In this work, we propose a MRI‐to‐CT transformer‐based improved denoising diffusion probabilistic model (MC‐IDDPM) to translate MRI into high‐quality sCT to facilitate radiation treatment planning.MethodsMC‐IDDPM implements diffusion processes with a shifted‐window transformer network to generate sCT from MRI. The proposed model consists of two processes: a forward process, which involves adding Gaussian noise to real CT scans to create noisy images, and a reverse process, in which a shifted‐window transformer V‐net (Swin‐Vnet) denoises the noisy CT scans conditioned on the MRI from the same patient to produce noise‐free CT scans. With an optimally trained Swin‐Vnet, the reverse diffusion process was used to generate noise‐free sCT scans matching MRI anatomy. We evaluated the proposed method by generating sCT from MRI on an institutional brain dataset and an institutional prostate dataset. Quantitative evaluations were conducted using several metrics, including Mean Absolute Error (MAE), Peak Signal‐to‐Noise Ratio (PSNR), Multi‐scale Structure Similarity Index (SSIM), and Normalized Cross Correlation (NCC). Dosimetry analyses were also performed, including comparisons of mean dose and target dose coverages for 95% and 99%.ResultsMC‐IDDPM generated brain sCTs with state‐of‐the‐art quantitative results with MAE 48.825 ± 21.491 HU, PSNR 26.491 ± 2.814 dB, SSIM 0.947 ± 0.032, and NCC 0.976 ± 0.019. For the prostate dataset: MAE 55.124 ± 9.414 HU, PSNR 28.708 ± 2.112 dB, SSIM 0.878 ± 0.040, and NCC 0.940 ± 0.039. MC‐IDDPM demonstrates a statistically significant improvement (with p < 0.05) in most metrics when compared to competing networks, for both brain and prostate synthetic CT. Dosimetry analyses indicated that the target dose coverage differences by using CT and sCT were within ± 0.34%.ConclusionsWe have developed and validated a novel approach for generating CT images from routine MRIs using a transformer‐based improved DDPM. This model effectively captures the complex relationship between CT and MRI images, allowing for robust and high‐quality synthetic CT images to be generated in a matter of minutes. This approach has the potential to greatly simplify the treatment planning process for radiation therapy by eliminating the need for additional CT scans, reducing the amount of time patients spend in treatment planning, and enhancing the accuracy of treatment delivery.

Synthetic CT imaging for PET monitoring in proton therapy: a simulation study

“…In the case of MAE, 75% of the counts of all slices of all patients are localized in HU values below 79 HU, as in Kurz et al (2019), Maspero et al (2020), Thummerer et al (2020). Similarly, in the PSNR case, most of the counts occupied the dB values from 15 up to 30 dB, comparable with Chen et al (2022), Zhang et al (2022), Pan et al (2023). In the DSC case, all the histograms are characterized by narrow peaks.…”

Dose-weighted proton linear energy transfer map generation using a deep learning framework

“…DL techniques have been extensively integrated into the field of medical physics. Their applications span various areas such as image synthesis [14,15], noise reduction [16,17],and mass density prediction [18][19][20]. Our proposed framework for generating synthetic LETd maps, derived from dose maps, encompasses two stages: a training stage and a test stage.…”

Image-domain material decomposition for dual-energy CT using a conditional diffusion model

“…The other is that the one-step material decomposition associated with the deep learning-based approaches is much more computationally efficient than the iterative optimization process. Recently, the diffusion model is emerging as a generative approach and gaining much attention in image synthesis [10,11], image translation [12][13][14][15], and image super resolution [16], due to its superior performance to other generative models like the generative adversarial model (GAN) [17]. However, there is still a lack of studies to prove its feasibility in material decomposition applications.…”