MultiRecon Project Completed

We are pleased to share that the MultiRecon project has now concluded.

The final report has been submitted to the ANR and will be made public soon.

Overview

MultiRecon developed machine-learning methods for multimodal medical imaging (PET/CT and PET/MRI). Our main objectives were:

  • Improve image quality by reducing noise
  • Combine complementary information from different modalities
  • Lower radiation dose for patients
  • Focus on unsupervised learning

Highlights

  1. Algorithms and Models
    • Multi-channel generative models for joint PET/CT, PET/MRI and cpectral CT
    • Inter-modal consistency loss for more reliable results
  2. Validation
    • Benchmarked on datasets from LaTIM, CREATIS, SHFJ and CHU Poitiers
    • Achieved sginificant noise reduction in most tests
  3. Publications
    • 6 papers published in peer-reviewed international journals
    • 4 preprints submitted to peer-reviewed international journals
  4. Software
    • Python packages for PET/CT and PET/MRI synergistic reconstruction
  5. Dissemination
    • Presentations at IEEE NSS MIC RTSD, IEEE ISBI and Eusipco
    • PhD defenses by Noel Jeffrey Pinton, Zhihan Wang and Valentin Gautier

Acknowledgements

We thank the French National Research Agency (ANR‑20‑CE45‑0020) and our partner institutions: LaTIM, CREATIS, BioMaps, SHFJ, and CHU Poitiers.

Next Steps

Ongoing efforts include:

  • Integrating our models into clinical workflows
  • Extending methods to other imaging modalities