Skip to content
Open Imaging Index
MRI Brain

crossMoDA

Cross-Modality Domain Adaptation Challenge (crossMoDA 2021, 2022 and 2023)

Unpaired contrast-enhanced T1 and high-resolution T2 brain MRI of patients with a vestibular schwannoma, released for a challenge on unsupervised domain adaptation from T1 to T2 for tumour and cochlea segmentation, with Koos grades added in 2022.

Go to dataset Edit Report

Overview

crossMoDA is a MICCAI challenge series (2021, 2022, 2023) run by King's College London on unsupervised cross-modality domain adaptation. Participants get annotated contrast-enhanced T1 (ceT1) scans and separate, unannotated high-resolution T2 (hrT2) scans of other patients, and must segment the vestibular schwannoma and both cochleas on hrT2. The 2022 edition added Koos grade classification; the 2023 edition split the tumour into intra- and extra-meatal parts and added routine surveillance scans from several UK hospitals. This entry counts only the training and validation data the challenge released; test sets were kept private.

Composition

The 2021 release holds the 242 London patients of Vestibular-Schwannoma-SEG, re-split into 105 annotated ceT1, 105 hrT2 training and 32 hrT2 validation scans. The 2022 release keeps these and adds 242 patients from Elisabeth-TweeSteden Hospital, Tilburg, with the same split, for 484 patients with one scan each. The 2023 release drops post-operative cases from London (196 remain) and adds 100 patients with 180 longitudinal scans from the UK multi-centre routine surveillance cohort (38 patients and 43 scans as annotated source, 47 and 105 as target, 15 and 32 for validation).

Acquisition

London scans come from a Siemens Avanto 1.5 T (MPRAGE ceT1; CISS hrT2, with TSE or FIESTA also named by the papers), Tilburg scans from a Philips Ingenia 1.5 T (3D-FFE ceT1, 3D-TSE hrT2). The UK surveillance scans span ten sites, 2006 to 2019, on Siemens, Philips, GE and Hitachi scanners at 1.0, 1.5 and 3 T. All images were de-identified, defaced and distributed as NIfTI.

Annotations

Tumours were contoured in Leksell GammaPlan by the treating neurosurgeon and physicist; radiology fellows added both cochleas in ITK-SNAP on T2. The 2022 release includes GIF parcellation masks for the source training scans and a CSV with the centre and Koos grade per source case, graded by two neurosurgeons with consensus review. In 2023 a radiologist sub-segmented the tumour into intra- and extra-meatal parts.

Known limitations

  • The London part is not new data; it repeats Vestibular-Schwannoma-SEG in NIfTI with extra cochlea labels.
  • Source and target scans are unpaired by design, so no patient has both contrasts in the release.
  • The 2022 Zenodo record lists CC BY 4.0 in its metadata but CC BY-NC-SA 4.0 in its description.
  • 2023 data required challenge registration and a request form.

Cohort

Aggregate numbers from the sources below. Bars are relative to the 484 subjects.

Contrast / sequence

Groups can overlap

  • T2-weighted 274 57%
  • T1-weighted, contrast enhanced 210 43%

Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

Field strength

  • 1.5 T 484 100%

Section 2.2.2 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

Country

  • United Kingdom 242 50%
  • Netherlands 242 50%

Section 3.2.1 in Dorent et al. 2023, Medical Image Analysis 83, 102628 (crossMoDA 2021 challenge paper, arXiv:2201.02831); Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

Country by split

Reported cross table. A dot marks a cell the source does not give.

TrainingValidation
United Kingdom21032
Netherlands21032

Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

Condition

Groups can overlap

  • Vestibular schwannoma 484 100%

Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

Scanner vendor

  • Siemens Healthineers 242 50%
  • Philips 242 50%

Section 2.2.2 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

Split

  • Training 420 87%
  • Validation 64 13%

Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

Contrast combinations

How many subjects have exactly each set of contrasts.

T1w_ceT2wSubjects with exactly this set
274
210

Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

Other reported numbers 4
Subjects · United Kingdom, T1-weighted, contrast enhanced 105Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)
Subjects · United Kingdom, T2-weighted 137Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)
Subjects · Netherlands, T1-weighted, contrast enhanced 105Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)
Subjects · Netherlands, T2-weighted 137Table 1 in Wijethilake, Dorent et al. 2025, crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023 (arXiv:2506.12006v3)

License and access

Our reading of the license, not legal advice. Before you use the data, read the original license and confirm that your use is allowed. We take no responsibility for how you use a dataset. Full disclaimer

Access
Open download

Download without an account

The 2021 (Zenodo 4662239) and 2022 (Zenodo 6504722) training and validation sets download without an account. The 2023 sets, which add the UK multi-centre routine surveillance scans, were distributed on Synapse (syn51236108) after challenge registration and a data access request form. Test sets of all editions stay private.

Access page

Different parts of the data carry different licenses. The summary on the right shows the most restrictive answer per rule.

For crossMoDA 2021 release on Zenodo (record 4662239), London scans only.

Creative Commons Attribution 4.0 International

Use, share and adapt the data for any purpose, including commercial use, as long as you credit the creators.

The 2021 challenge paper states the data was released under CC-BY-4.0, and the Zenodo record carries the same license.

Original license text Version read: 4.0 Checked 2026-10-07

What you can do

  • Yes
  • Yes
  • Yes
  • Yes
  • Yes

What you can share

  • Yes
  • Yes
  • Yes

What you must do

  • Yes
  • Share alike No
  • No
  • No
  • Manuscript review No
  • Release code No
  • Return results No
  • Delete after use No

Limits

  • No
  • Location limits No
For crossMoDA 2022 release on Zenodo (record 6504722) and the 2023 release on Synapse.

Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International

Non-commercial use only, with credit. Anything you share that is built on the data must use the same license.

The challenge website, the description of Zenodo record 6504722 and the 2021 to 2023 challenge paper state CC BY-NC-SA 4.0. The license field of the Zenodo record itself says CC BY 4.0; the stricter stated license is used here. The 2023 Synapse terms also require citing the 2021 challenge paper and two dataset papers.

Original license text Version read: 4.0 Checked 2026-10-07

What you can do

  • No
  • Not stated
  • Conditional
  • Yes
  • Yes

What you can share

  • Conditional
  • Conditional
  • Conditional

What you must do

  • Yes
  • Yes
  • No
  • No
  • Manuscript review No
  • Release code No
  • Return results No
  • Delete after use No

Limits

  • No
  • Location limits No

Citation

Dorent R, Kujawa A, Ivory M, et al. CrossMoDA 2021 challenge: Benchmark of cross-modality domain adaptation techniques for vestibular schwannoma and cochlea segmentation. Medical Image Analysis 83, 102628 (2023). https://doi.org/10.1016/j.media.2022.102628. Also cite: Shapey J, Kujawa A, Dorent R, et al. (2021). Segmentation of Vestibular Schwannoma from Magnetic Resonance Imaging: An Open Annotated Dataset and Baseline Algorithm [Data set]. The Cancer Imaging Archive. https://doi.org/10.7937/TCIA.9YTJ-5Q73

Sources

Every number on this page comes from one of these documents. Each chart names the table or page it is taken from. The raw numbers are in stats.csv.