Published January 14, 2025 | Version v1
Dataset Open

Brain Tumor Sequence Registration (BraTS-Reg) Challenge Dataset

Authors/Creators

  • 1. University of Pennsylvania
  • 2. ROR icon Emory University
  • 3. ROR icon Georgia Institute of Technology
  • 4. ROR icon Washington University in St. Louis
  • 5. ROR icon Santa Clara University
  • 6. ROR icon Technical University of Munich
  • 7. ROR icon Duke University
  • 8. ROR icon University of California, San Diego
  • 9. ROR icon Washington University in St. Louis School of Medicine
  • 10. USF Health Morsani College of Medicine
  • 11. ROR icon University of California, San Francisco
  • 12. EDMO icon Helmholtz Munich
  • 13. ROR icon Massachusetts Institute of Technology
  • 14. ROR icon University College London
  • 15. Univerisity of Zurich
  • 16. ROR icon Indiana University School of Medicine
  • 17. ROR icon University Hospital Regensburg

Description

Registration of longitudinal brain Magnetic Resonance Imaging (MRI) scans containing pathologies poses significant challenges due to substantial changes in tissue appearance. While general-purpose medical image registration techniques have advanced, they still lack the precision and reliability required for this complex task.

To address this gap, we organized the first-ever Brain Tumor Sequence Registration (BraTS-Reg) Challenge, which provides a public benchmark for deformable registration algorithms. The challenge focuses on establishing correspondences between pre-operative and follow-up scans of patients with diffuse brain gliomas. BraTS-Reg was conducted in conjunction with the IEEE International Symposium on Biomedical Imaging (ISBI) 2022 and the International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2022.

The challenge offers de-identified, multi-institutional, multi-parametric MRI scans, along with ground truth (GT) landmark points annotated by clinical experts at distinct anatomical locations across the temporal domain. To facilitate algorithm development, we are releasing the training data with GT annotations, while the validation data does not include GT annotations. During the testing phase, containerized algorithms were evaluated on hidden hold-out data.

BraTS-Reg serves as an active resource for research, with data and online evaluation tools available at https://bratsreg.github.io/. Quantitative evaluation and rankings during the challenge were based on key metrics, including Median Euclidean Error (MEE), Robustness, and the determinant of the Jacobian of the displacement field.

Files

BraTSReg_Training_Data.zip

Files (4.3 GB)

Name Size
md5:3a3026d6b8de4699b22e6edf0472bdd2
3.7 GB Preview Download
md5:d9ef410744dc93832b05f5e8763cc6d7
533.1 MB Preview Download

Additional details

Related works

Is described by
Preprint: 10.48550/arXiv.2112.06979 (DOI)

Funding

National Institutes of Health
R01NS042645
National Institutes of Health
U24CA189523
National Institutes of Health
T32EB001631
Radiological Society of North America
RR2011
Deutsche Forschungsgemeinschaft
GSC 81