cs.CVOct 27, 2026

MRI-to-CT Translation of Spinal Images via Anatomy-Guided Pseudo-Pairing and Adversarial Diffusion

Authors: Eduardo Nieto, Ibai Genua, Davide Scorza, Karen López-Linares

Organizations: Fundación Vicomtech, Basque Research and Technology Alliance (BRTA), Donostia - San Sebastián, Spain · R&D Department, Cyber Surgery, Donostia - San Sebastián, Spain · eHealth Group, Bioengineering Area, Biogipuzkoa Health Research Institute, Donostia - San Sebastián, Spain

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Magnetic Resonance Imaging (MRI) is routinely used to diagnose spinal disorders, but surgical planning often requires Computed Tomography (CT) for better visualization of vertebral bone anatomy. Synthesizing CT from MRI could avoid additional radiation exposure and streamline preoperative assessment. We propose a three-stage framework for spine MRI-to-CT synthesis. First, an anatomy-guided vertebral matching system identifies similar sagittal MRI and CT slices using spine segmentations and aligns them through rigid registration, creating pseudo-paired data from unpaired scans. Second, these pairs train an adversarial diffusion model for MRI-to-CT translation. Third, synthesized slices are assembled into 3D volumes and refined by a network that improves inter-slice continuity and structural coherence. The framework was trained on over 10,000 unpaired sagittal MRI and CT slices and evaluated on 1,911 slices from 83 MRI volumes spanning all spinal regions. Assessment combined distribution-based measures with vertebral segmentation metrics. Compared with diffusion alone, the proposed approach increased binary 3D vertebral Dice from 0.666 to 0.739. Volumetric post-processing reduced visible discontinuities between stacked slices and improved vertebral coherence in axial and coronal views. Combining anatomy-guided pseudo-pairing, adversarial diffusion and 3D refinement improves spine MRI-to-CT synthesis. Vertebral matching primarily enhanced anatomical alignment and fidelity, while volumetric refinement improved 3D consistency. These findings support further development of MRI-derived CT-like visualization for preoperative planning, particularly to preserve fine intensity details while maintaining volumetric coherence.

Figures & tables

Figure 1: Data flow diagram illustrating the full pipeline from raw data to the final dataset construction. Starting from 2844 CT scans and 560 MR scans, vertebral segmentations are obtained via TotalSegmentator, followed by manual filtering and region-based subvolume extraction. Subvolumes are then automatically filtered based on the quality of their segmentation masks, discarding those with anomalous or incomplete vertebral segmentations. Since the downstream translation model operates on 2D, sagittal slices are extracted from the subvolumes. Finally, the resulting slices are split into training, validation and test sets.
SplitRegionMRCTMR SlicesCT Slices
TrainCervical198340453210303
Cervical & Thoracic133342979984
Thoracic12302827082984
Thoracic & Lumbar1293248280
Lumbar2612412619685416
ValidationCervical6010713713228
Table 1: Dataset split summary after sagittal slice extraction. MR and CT columns indicate the number of subvolumes per modality, while MR Slices and CT Slices reflect the total number of 2D sagittal slices extracted from the subvolumes. The statistics are further stratified by spinal region. A vertebral region is considered present if it contains at least 50% of the possible vertebrae from that region (e.g 4 out of 7 cervical vertebrae).
Figure 2: Example of MR (left) and CT (right) patches extraction after preprocessing. From each scan, multiple overlapping patches covering different spinal regions are extracted, as shown by the individual patch examples on the right of each scan. Unlike CT, where redundant patches from the same scan are allowed, MR patches are filtered to avoid patches with similar spine information, resulting in a significantly smaller number of extracted MR patches overall.
DatasetMR-CT pseudo-pairsMean DSCDiffering vertebrae
   01>1>1
Train113190.6374645948555
Validation35550.6340182517282
Test19110.61809759288
Table 2: Vertebral matching system results for the different datasets. The mean DSC value represents the average overlap between the MR and CT spines in the generated pseudo-pairs. The differing vertebrae indicates the number of vertebral mismatches between the sagittal slices of every generated pseudo-pair, with 0 meaning both slices have the same vertebrae, 1 meaning there is at least one vertebra that is not present in both slices and ¿1 meaning there is more than 1 vertebrae that is not present in both slices.
Figure 3: Vertebral matching system overview. Figure illustrates how every MR sagittal slice is paired with a slice of the CT corresponding set, trying to maximize the spine similarity and overlap between each pair through a rigid registration process. Once paired, both slices are cropped to avoid empty pairings resulting from excessive rotations or shifts from rigid registration.
Figure 4: Vertebral matching system empirical evaluation on the test dataset. The top row shows the obtained weighted DSC, execution time and matching efficiency (DSC increase per second of execution) as a function of the number of evaluated candidates. The bottom plot applies the elbow method to find the optimal trade-off between patch spinal similarity (DSC) and computational cost, identifying 50 candidates as the optimal selection.
Figure 5: Overview of the post-processing net architecture. The ResUNet generator takes a synthetic CT as input and produces a corrected CT output, trying to increase its volumetric consistency and coherence. Residual skip connections concatenate encoder and decoder feature maps, enabling multi-scale feature fusion and improved anatomical detail preservation. The PatchGAN discriminator classifies local regions of the output as real or fake to provide adversarial supervision.
Figure 6: Example of the simulated discontinuity artifacts generated through the vertebral matching system a on real CT volume.
Figure 7: Progressive qualitative impact of each pipeline stage. Rows show representative and challenging examples across different spinal regions.
 Global realismVertebral fidelity
MethodrMMD ↓Full FID ↓TS 3D DSC (bin.)↑TS 3D DSC ↑Vertebral FID ↓EMD ↓JSD ↓
SynDiff baseline0.13577.840.666±0.070.435±0.1741.57111.2±61.10.308±0.13
+ pseudo-matching0.05380.880.739±0.040.564±0.1533.6485.7±440.267±0.13
+ 3D post-processing0.07487.410.732±0.040.575±0.13350.3776.8±63.20.213±0.14
Table 3: Incremental contribution of each pipeline stage. Each row adds a component on top of the previous one. Global realism metrics evaluate the overall image quality, while vertebral fidelity metrics asses the anatomical consistency of the generated CT. TS 3D DSC reports the multiclass Dice per vertebra, while TS 3D DSC (bin.) treats all vertebrae as a single foreground class. Values are reported in a mean ±\pm and standard deviation (when available) format computed over the 83 volumes of the test set. ↑\uparrow indicates higher is better, ↓\downarrow lower is better.
Figure 8: Vertebra-wise intensity fidelity before and after 3D refinement.
Figure 9: Volumetric vertebral coherence before and after 3D post-processing, seen through the vertebral segmentations in the different anatomical planes. The left column shows the MR ground truth segmentation, the middle column the synthetic CT segmentation obtained with the 2D translation model and the vertebral matching system approach and the right column shows the result after applying the 3D post-processing network.
More in this topic

Medical Image Synthesis

Medical image synthesis creates, translates, reconstructs, and edits clinical images to fill missing data and support analysis.

See all 310 papers

Explore similar work

CardsList
Mar 30, 2026

MRI-to-CT synthesis using drifting models

Accurate MRI-to-CT synthesis could enable MR-only pelvic workflows by providing CT-like images with bone details while avoiding additional ionizing radiation. In this work, we investigate recently proposed drifting models for synthesizing pelvis CT images from MRI and benchmark them against convolutional neural networks (UNet, VAE), a generative adversarial network (WGAN-GP), a physics-inspired probabilistic model (PPFM), and diffusion-based methods (FastDDPM, DDIM, DDPM). Experiments are performed on two complementary datasets: Gold Atlas Male Pelvis and the SynthRAD2023 pelvis subset. Image fidelity and structural consistency are evaluated with SSIM, PSNR, and RMSE, complemented by qualitative assessment of anatomically critical regions such as cortical bone and pelvic soft-tissue interfaces. Across both datasets, the proposed drifting model achieves high SSIM and PSNR and low RMSE, surpassing strong diffusion baselines and conventional CNN-, VAE-, GAN-, and PPFM-based methods. Visual inspection shows sharper cortical bone edges, improved depiction of sacral and femoral head geometry, and reduced artifacts or over-smoothing, particularly at bone-air-soft tissue boundaries. Moreover, the drifting model attains these gains with one-step inference and inference times on the order of milliseconds, yielding a more favorable accuracy-efficiency trade-off than iterative diffusion sampling while remaining competitive in image quality. These findings suggest that drifting models are a promising direction for fast, high-quality pelvic synthetic CT generation from MRI and warrant further investigation for downstream applications such as MRI-only radiotherapy planning and PET/MR attenuation correction.

Qing Lyu, Jianxu Wang, Jeremy Hudson +2

Department of Radiology and Biomedical Imaging, Yale School of Medicine, New Haven, CT · Department of Biomedical Engineering, Rensselaer Polytechnic Institute, Troy, NY

Medical Image SynthesisDrifting ModelsMagnetic Resonance ImagingComputed Tomography
AskPDF
Mar 13, 2026

A Systematic Benchmark of GAN Architectures for MRI-to-CT Synthesis

The translation from Magnetic resonance imaging (MRI) to Computed tomography (CT) has been proposed as an effective solution to facilitate MRI-only clinical workflows while limiting exposure to ionizing radiation. Although numerous Generative Adversarial Network (GAN) architectures have been proposed for MRI-to-CT translation, systematic and fair comparisons across heterogeneous models remain limited. We present a comprehensive benchmark of ten GAN architectures evaluated on the SynthRAD2025 dataset across three anatomical districts (abdomen, thorax, head-and-neck). All models were trained under a unified validation protocol with identical preprocessing and optimization settings. Performance was assessed using complementary metrics capturing voxel-wise accuracy, structural fidelity, perceptual quality, and distribution-level realism, alongside an analysis of computational complexity. Supervised Paired models consistently outperformed Unpaired approaches, confirming the importance of voxel-wise supervision. Pix2Pix achieved the most balanced performance across districts while maintaining a favorable quality-to-complexity trade-off. Multi-district training improved structural robustness, whereas intra-district training maximized voxel-wise fidelity. This benchmark provides quantitative and computational guidance for model selection in MRI-only radiotherapy workflows and establishes a reproducible framework for future comparative studies. To ensure the reproducibility of our experiments we make our code public, together with the overall results, at the following link:https://github.com/arco-group/MRI_TO_CT.git

Alessandro Pesci, Valerio Guarrasi, Marco Alì +2

Unit of Artificial Intelligence and Computer Systems, Department of Engineering, Università Campus Bio-Medico di Roma, Italy · CDI Centro Diagnostico Italiano S.p.A., Milan, Italy · University of Milan-Bicocca, Milan, Italy +1

Medical Image SynthesisGenerative Adversarial NetworksMedical Image Benchmarks
AskPDF
Sep 12, 2026

TotalSynth: Robust Whole-Body Synthetic CT from MRI and CBCT

Purpose: To develop and evaluate TotalSynth, a reusable pretrained framework for whole-body synthetic CT (sCT) generation from MRI and cone-beam CT (CBCT) images. Materials and Methods: In this retrospective technical study, the dataset was assembled between 2020 and 2026 from SynthRAD challenge data, four prostate cohorts, and BIC-MAC. After registration quality control, 1450 of 1800 public challenge pairs were retained; 350 were excluded for insufficient registration quality or major source/CT mismatch. The corpus also included 84 additional prostate MRI/CT and 60 external BIC-MAC MRI/CT cases. Three 5-fold model families were evaluated with image-domain, anatomy-aware, registration-based, and uncertainty metrics. Age and sex were not consistently available across public datasets. Results: The released MRI-to-CT model achieved an overall MAE of 67.49 HU, SSIM of 0.920, and PSNR of 29.28 dB. The CBCT-to-CT model achieved an overall MAE of 53.55 HU, SSIM of 0.939, and PSNR of 32.09 dB. The unified model maintained similar performance on MRI inputs (MAE, 67.68 HU) and CBCT inputs (MAE, 54.22 HU). On external BIC-MAC data, MRI-to-CT MAE was 100.91 HU without fine-tuning and 62.21 HU after fine-tuning. Conclusion: TotalSynth provides reusable MRI- and CBCT-based CT synthesis models with broad anatomical coverage, while external evaluation highlights the need for local validation and optional fine-tuning under domain shift.

Valentin Boussot, Cedric Hemon, Anais Barateau +3

Univ. Rennes, CLCC Eugène Marquis, INSERM, LTSI - UMR 1099 F-35000 Rennes, France

Medical Image SynthesisComputed Tomography
AskPDF