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

Mohammed A. Mahdi; Mohammed Al-Shalabi; Ehab T. Alnfrawy; Reda Elbarougy; Muhammad Usman Hadi; Rao Faizan Ali

doi:10.3390/diagnostics15233010

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

Mohammed A. Mahdi, Mohammed Al-Shalabi, Ehab T. Alnfrawy, Reda Elbarougy, Muhammad Usman Hadi, Rao Faizan Ali

University of Kent

Research output: Contribution to journal › Article › peer-review

1 Downloads (Pure)

Abstract

Background: The clinical imperative to reduce patient ionizing radiation exposure during diagnosis and treatment planning necessitates robust, high-fidelity synthetic imaging solutions. Current cross-modal synthesis techniques, primarily based on GANs and deterministic CNNs, exhibit instability and critical errors in modeling high-contrast tissues, thereby hindering their reliability for safety-critical applications such as radiotherapy. Objectives: Our primary objective was to develop a stable, high accuracy framework for 3D Magnetic Resonance Imaging (MRI) to Computed Tomography (CT) synthesis capable of generating clinically equivalent synthetic CTs (sCTs) across multiple anatomical sites. Methods: We introduce a novel 3D Latent Diffusion Model (3DLDM) that operates in a compressed latent space, mitigating the computational burden of 3D diffusion while leveraging the stability of the denoising objective. Results: Across the Head & Neck, Thorax, and Abdomen, the 3DLDM achieved a Mean Absolute Error (MAE) of 56.44 Hounsfield Units (HU). This result demonstrates a significant 3.63% reduction in overall error compared to the strongest adversarial baseline, CycleGAN (MAE = 60.07 HU, p < 0.05), a 10.76% reduction compared to NNUNet (MAE = 67.20 HU, p < 0.01), and a 20.79% reduction compared to the transformer-based SwinUNeTr (MAE = 77.23 HU, p < 0.0001). The model also achieved the highest structural similarity (SSIM = 0.885 ± 0.031), significantly exceeding SwinUNeTr ( p < 0.0001), NNUNet ( p < 0.01), and Pix2Pix ( p < 0.0001). Likewise, the 3D-LDM achieved the highest peak signal-to-noise ratio (PSNR = 29.73 ± 1.60 dB), with statistically significant gains over CycleGAN ( p < 0.01), NNUNet ( p < 0.001), and SwinUNeTr ( p < 0.0001). Conclusions: This work validates a scalable, accurate approach for volumetric synthesis, positioning the 3DLDM to enable MR-only radiotherapy planning and accelerate radiation-free multi-modal imaging in the clinic.

Original language	English
Article number	3010
Pages (from-to)	1-18
Number of pages	18
Journal	Diagnostics
Volume	15
Issue number	23
Early online date	26 Nov 2025
DOIs	https://doi.org/10.3390/diagnostics15233010
Publication status	Published online - 26 Nov 2025

Bibliographical note

Publisher Copyright:
© 2025 by the authors.

Data Access Statement

The data supporting the findings of this study are publicly available on Zenodo (https://doi.org/10.5281/zenodo.7260705) under the SynthRAD2023 collection (Accessed: 19 January 2025). No additional data were generated or analyzed beyond those reported in this study.

Funding

This research has been funded by Scientific Research Deanship at University of Ha’il-Saudi Arabia through project number RG-24 182.

Keywords

latent diffusion models
medical image synthesis
MRI-to-CT tanslation
3D volumetric imaging
generative models
synthetic CT
radiotherapy planning
3d Volumetric Imaging
Mri-to-ct Translation
Latent Diffusion Models
Medical Image Synthesis
Generative Models
Synthetic Ct
Radiotherapy Planning

Access to Document

10.3390/diagnostics15233010

diagnostics-15-03010-v2.cleanedFinal published version, 2.17 MBLicence: CC BY

Cite this

@article{0dec5195a8a4498186ee7899d557c5d4,

title = "3D Latent Diffusion Model for MR-Only Radiotherapy: Accurate and Consistent Synthetic CT Generation",

abstract = " Background: The clinical imperative to reduce patient ionizing radiation exposure during diagnosis and treatment planning necessitates robust, high-fidelity synthetic imaging solutions. Current cross-modal synthesis techniques, primarily based on GANs and deterministic CNNs, exhibit instability and critical errors in modeling high-contrast tissues, thereby hindering their reliability for safety-critical applications such as radiotherapy. Objectives: Our primary objective was to develop a stable, high accuracy framework for 3D Magnetic Resonance Imaging (MRI) to Computed Tomography (CT) synthesis capable of generating clinically equivalent synthetic CTs (sCTs) across multiple anatomical sites. Methods: We introduce a novel 3D Latent Diffusion Model (3DLDM) that operates in a compressed latent space, mitigating the computational burden of 3D diffusion while leveraging the stability of the denoising objective. Results: Across the Head \& Neck, Thorax, and Abdomen, the 3DLDM achieved a Mean Absolute Error (MAE) of 56.44 Hounsfield Units (HU). This result demonstrates a significant 3.63\% reduction in overall error compared to the strongest adversarial baseline, CycleGAN (MAE = 60.07 HU, p < 0.05), a 10.76\% reduction compared to NNUNet (MAE = 67.20 HU, p < 0.01), and a 20.79\% reduction compared to the transformer-based SwinUNeTr (MAE = 77.23 HU, p < 0.0001). The model also achieved the highest structural similarity (SSIM = 0.885 ± 0.031), significantly exceeding SwinUNeTr ( p < 0.0001), NNUNet ( p < 0.01), and Pix2Pix ( p < 0.0001). Likewise, the 3D-LDM achieved the highest peak signal-to-noise ratio (PSNR = 29.73 ± 1.60 dB), with statistically significant gains over CycleGAN ( p < 0.01), NNUNet ( p < 0.001), and SwinUNeTr ( p < 0.0001). Conclusions: This work validates a scalable, accurate approach for volumetric synthesis, positioning the 3DLDM to enable MR-only radiotherapy planning and accelerate radiation-free multi-modal imaging in the clinic. ",

keywords = "latent diffusion models, medical image synthesis, MRI-to-CT tanslation, 3D volumetric imaging, generative models, synthetic CT, radiotherapy planning, 3d Volumetric Imaging, Mri-to-ct Translation, Latent Diffusion Models, Medical Image Synthesis, Generative Models, Synthetic Ct, Radiotherapy Planning",

author = "Mahdi, \{Mohammed A.\} and Mohammed Al-Shalabi and Alnfrawy, \{Ehab T.\} and Reda Elbarougy and Hadi, \{Muhammad Usman\} and Ali, \{Rao Faizan\}",

note = "Publisher Copyright: {\textcopyright} 2025 by the authors.",

year = "2025",

month = nov,

day = "26",

doi = "10.3390/diagnostics15233010",

language = "English",

volume = "15",

pages = "1--18",

journal = "Diagnostics",

issn = "2075-4418",

publisher = "MDPI",

number = "23",

}

TY - JOUR

T1 - 3D Latent Diffusion Model for MR-Only Radiotherapy: Accurate and Consistent Synthetic CT Generation

AU - Mahdi, Mohammed A.

AU - Al-Shalabi, Mohammed

AU - Alnfrawy, Ehab T.

AU - Elbarougy, Reda

AU - Hadi, Muhammad Usman

AU - Ali, Rao Faizan

PY - 2025/11/26

Y1 - 2025/11/26

N2 - Background: The clinical imperative to reduce patient ionizing radiation exposure during diagnosis and treatment planning necessitates robust, high-fidelity synthetic imaging solutions. Current cross-modal synthesis techniques, primarily based on GANs and deterministic CNNs, exhibit instability and critical errors in modeling high-contrast tissues, thereby hindering their reliability for safety-critical applications such as radiotherapy. Objectives: Our primary objective was to develop a stable, high accuracy framework for 3D Magnetic Resonance Imaging (MRI) to Computed Tomography (CT) synthesis capable of generating clinically equivalent synthetic CTs (sCTs) across multiple anatomical sites. Methods: We introduce a novel 3D Latent Diffusion Model (3DLDM) that operates in a compressed latent space, mitigating the computational burden of 3D diffusion while leveraging the stability of the denoising objective. Results: Across the Head & Neck, Thorax, and Abdomen, the 3DLDM achieved a Mean Absolute Error (MAE) of 56.44 Hounsfield Units (HU). This result demonstrates a significant 3.63% reduction in overall error compared to the strongest adversarial baseline, CycleGAN (MAE = 60.07 HU, p < 0.05), a 10.76% reduction compared to NNUNet (MAE = 67.20 HU, p < 0.01), and a 20.79% reduction compared to the transformer-based SwinUNeTr (MAE = 77.23 HU, p < 0.0001). The model also achieved the highest structural similarity (SSIM = 0.885 ± 0.031), significantly exceeding SwinUNeTr ( p < 0.0001), NNUNet ( p < 0.01), and Pix2Pix ( p < 0.0001). Likewise, the 3D-LDM achieved the highest peak signal-to-noise ratio (PSNR = 29.73 ± 1.60 dB), with statistically significant gains over CycleGAN ( p < 0.01), NNUNet ( p < 0.001), and SwinUNeTr ( p < 0.0001). Conclusions: This work validates a scalable, accurate approach for volumetric synthesis, positioning the 3DLDM to enable MR-only radiotherapy planning and accelerate radiation-free multi-modal imaging in the clinic.

AB - Background: The clinical imperative to reduce patient ionizing radiation exposure during diagnosis and treatment planning necessitates robust, high-fidelity synthetic imaging solutions. Current cross-modal synthesis techniques, primarily based on GANs and deterministic CNNs, exhibit instability and critical errors in modeling high-contrast tissues, thereby hindering their reliability for safety-critical applications such as radiotherapy. Objectives: Our primary objective was to develop a stable, high accuracy framework for 3D Magnetic Resonance Imaging (MRI) to Computed Tomography (CT) synthesis capable of generating clinically equivalent synthetic CTs (sCTs) across multiple anatomical sites. Methods: We introduce a novel 3D Latent Diffusion Model (3DLDM) that operates in a compressed latent space, mitigating the computational burden of 3D diffusion while leveraging the stability of the denoising objective. Results: Across the Head & Neck, Thorax, and Abdomen, the 3DLDM achieved a Mean Absolute Error (MAE) of 56.44 Hounsfield Units (HU). This result demonstrates a significant 3.63% reduction in overall error compared to the strongest adversarial baseline, CycleGAN (MAE = 60.07 HU, p < 0.05), a 10.76% reduction compared to NNUNet (MAE = 67.20 HU, p < 0.01), and a 20.79% reduction compared to the transformer-based SwinUNeTr (MAE = 77.23 HU, p < 0.0001). The model also achieved the highest structural similarity (SSIM = 0.885 ± 0.031), significantly exceeding SwinUNeTr ( p < 0.0001), NNUNet ( p < 0.01), and Pix2Pix ( p < 0.0001). Likewise, the 3D-LDM achieved the highest peak signal-to-noise ratio (PSNR = 29.73 ± 1.60 dB), with statistically significant gains over CycleGAN ( p < 0.01), NNUNet ( p < 0.001), and SwinUNeTr ( p < 0.0001). Conclusions: This work validates a scalable, accurate approach for volumetric synthesis, positioning the 3DLDM to enable MR-only radiotherapy planning and accelerate radiation-free multi-modal imaging in the clinic.

KW - latent diffusion models

KW - medical image synthesis

KW - MRI-to-CT tanslation

KW - 3D volumetric imaging

KW - generative models

KW - synthetic CT

KW - radiotherapy planning

KW - 3d Volumetric Imaging

KW - Mri-to-ct Translation

KW - Latent Diffusion Models

KW - Medical Image Synthesis

KW - Generative Models

KW - Synthetic Ct

KW - Radiotherapy Planning

UR - https://www.scopus.com/pages/publications/105024496085

U2 - 10.3390/diagnostics15233010

DO - 10.3390/diagnostics15233010

M3 - Article

C2 - 41374391

SN - 2075-4418

VL - 15

SP - 1

EP - 18

JO - Diagnostics

JF - Diagnostics

IS - 23

M1 - 3010

ER -

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

Abstract

Bibliographical note

Data Access Statement

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this