Image annotation tools are the software a team uses to draw boxes, polygons, and masks on images — including DICOM / NIfTI — so a model has labels to train on. This page is a tools list. It is not annotation-as-a-service, not a techniques explainer, and not a ranked “12 best 2025.”
If you meant how you label (box vs polygon vs landmark) → medical image annotation. If you meant someone else drawing the labels → medical image annotation services. If you meant datasets plus annotation in one hub → datasets & annotation hub.
PYCAD builds custom web DICOM viewers and the annotation → model → deploy path. It does not ship a public annotation SaaS to rank against Labelbox. A previous version of this page listed a “free PYCAD annotator” as #1; that URL 404s and the post is trashed. It is not on this list.
What to pick on (not a trophy)
| Criterion | Why it matters on medical images |
|---|---|
| Native DICOM / NIfTI / 3D | A JPEG box tool will not window-level a CT or keep the third dimension. If the file is a series, the tool has to know that. |
| Where it runs | PHI often cannot sit on a US vendor cloud. On-prem / VPC / air-gap is a gate, not a preference. |
| QA | Consensus, review stages, role-based access. One intern with a brush is not a ground-truth process. |
| Open vs paid | CVAT and Label Studio you can host. Labelbox / V7 / Encord you buy. Neither is “more medical.” |
| Pre-label | A model-in-the-loop (SAM, your own checkpoint) is how large sets get done. It is not a substitute for a clinician review on the hard cases. |
Tools that teams actually use
No rank. No stars. Medical-native is called out; general computer-vision platforms are listed as that.
| Tool | What it is | Medical-useful bit | Catch |
|---|---|---|---|
| Labelbox | Commercial data platform (catalog → annotate → model) | DICOM tools, polygons / masks, consensus workflows | Usage-based pricing; watch the bill on a large CT set. labelbox.com |
| V7 Darwin | Commercial, strong on medical and video | DICOM-native rendering, autoML-assisted segmentation, VPC option | Enterprise quote. v7labs.com |
| Encord | Commercial, active-learning / eval in the same shell | HIPAA / SOC 2 claims, on-prem and VPC | Sales-gated pricing; heavy if you only need boxes. encord.com |
| SuperAnnotate | Commercial, customizable editors | Builder for odd schemas (multi-layer organ, linked studies) | Orchestration is extra setup. superannotate.com |
| Supervisely | Commercial, modules for video / 3D / DICOM | 3D and DICOM add-ons, anonymization helpers | Modular price in EUR; the medical bits are often add-ons. supervisely.com |
| Dataloop | Commercial “data OS” | SAM-style pre-label, pipeline SDK | General CV that also does medical; GIS / LiDAR are not your CT problem. dataloop.ai |
| CVAT | Open source + CVAT.ai cloud | You can host it; interpolations and model helpers | Self-host is an IT job. Cloud is per-user. cvat.ai |
| Label Studio (HumanSignal) | Open-source core + paid cloud | On-prem when PHI cannot leave; XML-configured UI | Medical viewing is what you configure, not a built-in PACS. labelstud.io |
| Roboflow | End-to-end general CV (annotate → train → host) | Fast iteration, versioned datasets | Not a DICOM workstation. Fine for 2D photos (derm, wound); clumsy for a CT series. roboflow.com |
| Scale AI | Data engine + managed workforce | You can bring your own labelers or buy theirs | General, not medical-native. Enterprise is a sales cycle. scale.com |
| SageMaker Ground Truth | AWS managed labeling | Sits on S3 / IAM if you already live in AWS | Object-priced; UI is less of a medical viewer than V7 / Labelbox. AWS Ground Truth |
3D Slicer and ITK-SNAP are not on this table because they are viewers / research tools that happen to export a labelmap, not annotation platforms with a worklist. They still win for a one-off organ mask. See visualize / annotate DICOM & NIfTI.
A sane way to choose
- Name the file. 2D photo → almost anything. CT/MRI series → discard tools that only do JPEG.
- Name the leave-the-building rule. If PHI cannot go to a SaaS, you are on CVAT / Label Studio / an on-prem SKU. Full stop.
- Name the QA. Dual-read on the rare classes. One pass on the obvious background.
- Do not buy a platform to annotate fifty studies. Slicer is enough. Buy a platform when you have a production set and more than one labeler.
What this page is not
- Not a PYCAD annotation product. The free-tool URL 404s. Do not “try PYCAD” as item 12.
- Not the techniques article (5857) and not the services page (8005).
- Not HIPAA-as-a-feature-list from a document-sharing blog. Dropped documind.chat.
- Not Outrank screenshots or a comparison matrix with invented stars and trophy emoji. Dropped.
If the work is a custom viewer or a model that needs those labels, that is the imaging piece. Case studies.