TomasGuija/RCTSynth

SynthRCT: Scalable Conditional Deformation Synthesis for Synthetic Repeat CT Generation

This model is presented in the paper SynthRCT: Scalable Conditional Deformation Synthesis for Synthetic Repeat CT Generation.

RCTSynth generates synthetic repeat thoracic CT images by learning anatomical deformations. Given a source CT volume, the model predicts a stationary velocity field, integrates it into a dense displacement field, and warps the source image.

The original model uses a conditional variational registration architecture with a full-volume anatomy-conditioned prior and a memory-aware slab decoder. It supports patient-specific deformation sampling, latent-space traversal, and the export of warped CT volumes, stationary velocity fields, and displacement fields.

RCTSynth has also been extended with phase-conditioned respiratory motion synthesis. Given a CT at a fixed reference phase and a scalar target phase s ∈ [0, 1), the extended model synthesizes the corresponding respiratory state. The target phase conditions both the full-volume prior encoder and the local slab decoder. In this proof-of-concept formulation, the posterior path is not used and the conditioned prior mean is decoded directly. Inference is therefore deterministic.

Available checkpoints

Checkpoint	Description	Training data
rctsynth.ckpt	Original variational deformation-synthesis model	DIR-Lab 4DCT
phase_conditioned.ckpt	Deterministic phase-conditioned respiratory-motion model	MONAI Physio TCIA-4DLung tutorial subset

The original model checkpoint was trained on paired thoracic 4DCT images from the DIR-Lab 4DCT dataset, with case ID 0 reserved for validation. The training data are not redistributed here and remain subject to the dataset provider's terms.

The phase-conditioned checkpoint was trained using ten respiratory phases per subject, following the setup used by MONAI Physio. The phase approximately 70% through the respiratory cycle was used as the fixed source state, while the scalar target phase selected the state to synthesize.

License

The RCTSynth model weights are released under the MIT License. This license does not grant rights to the DIR-Lab training data or other third-party datasets.

Citation

@InProceedings{GuiTom_SynthRCT_MICCAISAT2026,
  author    = {Guija-Valiente, Tomás and Malpica, Norberto and Rodriguez-Gonzalez, Blanca},
  title     = {SynthRCT: Scalable Conditional Deformation Synthesis for Synthetic Repeat CT Generation},
  booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
  year      = {2026},
  publisher = {Springer Nature Switzerland},
  volume    = {LNCS 17274},
  month     = {pending},
  pages     = {pending}
}

References

1. Castillo, E., Castillo, R., Martinez, J., Shenoy, M., & Guerrero, T. (2010). Four-dimensional deformable image registration using trajectory modeling. Physics in Medicine and Biology, 55(1), 305–327. https://doi.org/10.1088/0031-9155/55/1/018.

2. Castillo, R., Castillo, E., Guerra, R., Johnson, V. E., McPhail, T., Garg, A. K., & Guerrero, T. (2009). A framework for evaluation of deformable image registration spatial accuracy using large landmark point sets. Physics in Medicine and Biology, 54(7), 1849–1870. https://doi.org/10.1088/0031-9155/54/7/001.

3. Hugo, G. D., Weiss, E., Sleeman, W. C., Balik, S., Keall, P. J., Lu, J., & Williamson, J. F. (2017). A longitudinal four-dimensional computed tomography and cone beam computed tomography dataset for image-guided radiation therapy research in lung cancer. Medical Physics, 44(2), 762–771. https://doi.org/10.1002/mp.12059

4. Hugo, G. D., Weiss, E., Sleeman, W. C., Balik, S., Keall, P. J., Lu, J., & Williamson, J. F. (2016). Data from 4D Lung Imaging of NSCLC Patients (4D-Lung). The Cancer Imaging Archive. https://doi.org/10.7937/K9/TCIA.2016.ELN8YGLE