Open data inside the workspace

Public CT and MRI datasets for segmentation practice

MedSeg includes a public dataset browser for real medical imaging datasets. Copy cases into a project, inspect reference masks when available, run models, and practice training without preparing every file yourself.

Browse after sign-in Train on corrected masks

CT MRI Pathology datasets Reference masks where available

MedSeg showing medical imaging masks that can be copied from public datasets. — Copied public cases behave like normal project series and can be segmented, edited, measured, and exported.

Watch public datasets in MedSeg

This demo shows how to start from Home, open Public datasets, select cases, copy them into a project, and open the copied cases in the MedSeg editor.

Open the video on YouTube

Browse the catalog

Every dataset has its own page with demo cases that open directly in the browser editor — no account needed.

TotalSegmentator (v2)

1228 real clinical CT scans with 117-structure ground-truth segmentations.

CT 1,228 cases CC-BY-4.0 labelled

AMOS22 — Abdominal Multi-Organ

500 CT + 100 MRI scans, 15 abdominal organs, multi-center / multi-vendor / multi-phase.

CT, MR 600 cases CC-BY-4.0 labelled

ReMIND — Brain tumor multi-modal MR (raw, with iUS)

114 brain tumor patients with pre/post-op multi-modal MR + intra-operative ultrasound. NOT skull-stripped.

MR, US 990 cases CC-BY-4.0 labelled

LIDC-IDRI — Lung CT with nodule pathology

1010 lung CT scans from 4 institutions, with marked-up nodules from 4 radiologists. Real clinical, multi-vendor.

CT 1,010 cases CC-BY-3.0 labelled

BHSD — Brain CT Intracranial Hemorrhage

Reconstructed BHSD head CT volumes: all labeled/unlabeled source files included; axial images preserved while AP slice d

CT 2,174 cases MIT labelled

HCC-TACE-Seg — Liver cancer multi-phase CT

QC-reimported HCC-TACE-Seg CT: TCIA DICOM series are split by within-series acquisition/stack metadata so multiphase sca

CT 628 cases CC-BY-3.0

TotalSegmentator — Pleural + Pericardial Effusion

600 CT cases with manual segmentations of pleural + pericardial effusions. Real clinical fluid pathology.

CT 600 cases CC-BY-4.0 labelled

Head CTA Aneurysm - DICOM with STL annotations

172 anonymized contrast-enhanced head CTA DICOM series from three clinical centers: 82 aneurysm cases and 90 controls.

CT 172 cases CC-BY-4.0 labelled

TotalSegmentator — Liver Lesions (CT)

842 CT cases with manual liver-lesion segmentations. Real clinical oncology cohort.

CT 842 cases CC-BY-4.0 labelled

Vestibular-Schwannoma-MC-RC - routine multi-center MR

160 vestibular schwannoma patients, 427 timepoints and 487 routine clinical MR DICOM series from 10 UK medical sites.

MR 487 cases CC-BY-4.0 labelled

TotalSegmentator — Liver Lesions (mixed MR + CT)

MR cases with manual liver-lesion segmentations. Multi-sequence routine clinical MRI. (mixed: 429 CT + 321 MR despite th

CT, MR 750 cases CC-BY-4.0 labelled

TotalSegmentator — Hip Implant

71 CT cases with manual hip-implant segmentations. Post-arthroplasty patients.

CT 71 cases CC-BY-4.0 labelled

TotalSegmentator — Liver Couinaud Segments (MR)

Manual segmentation of the 9 Couinaud liver segments — anatomical reference for surgery planning.

MR 59 cases CC-BY-4.0 labelled

Medical Decathlon — PROSTATE

Prostate transitional zone and peripheral zone segmentation

ADC, T2 64 cases CC-BY-SA-4.0 labelled

Dataset types currently represented

The catalog changes over time, but the current direction is practical: real image series, clear licenses, and clinically relevant anatomy/pathology.

General anatomy

TotalSegmentator, AMOS22, and Medical Decathlon-style datasets for organs and regional anatomy.

Pathology

LIDC-IDRI lung nodules, BHSD hemorrhage CT, ReMIND brain tumors, HCC-TACE-Seg liver cancer, and related lesion datasets.

Reference labels

Many datasets include reference masks that can be copied, inspected, corrected, or used as initial training data.

How public datasets help acquisition and onboarding

Users often want to see whether a segmentation tool works before uploading private data. Public cases solve that.

Open the dataset browser.
Filter by modality, anatomy, pathology, and free-text search.
Preview a case.
Check the scan, metadata, tags, and whether a reference mask exists.
Copy the case into a project.
The image is linked into your project without rewriting the underlying public series.
Run a workflow.
Try TotalSegmentator, nnInteractive, VoxTell, manual editing, or custom nnU-Net training.

License awareness

Open data is not the same as no obligations. MedSeg surfaces license metadata, but users still need to follow the source dataset terms.

License pattern	What to remember
CC BY	Attribution is required. Commercial use may be allowed by the dataset license.
CC BY-SA	Attribution plus ShareAlike obligations. Check downstream model-sharing plans carefully.
MIT	Permissive, but still keep source notices and dataset-specific documentation.
Model checkpoints	Some models, such as VoxTell checkpoints, can have non-commercial terms even if a dataset is commercial-use allowed.

Use public data responsibly

Do not mix up dataset license, model license, and clinical validation. A dataset can be open, a model can still have separate restrictions, and MedSeg remains a research tool.

Custom nnU-Net training VoxTell license note MedSeg Terms