Radiologist-led · CT & MRI
Ground truth your model can be judged against.
Annotomy gives medical imaging AI teams expert annotation, reference-standard labels and independent model evaluation. Every case is read by radiologists, not general labellers.
Services
Three things, done properly.
We work with teams building detection, segmentation and triage models on CT and MRI, from first training set to regulatory submission.
Expert annotation
Pixel- and voxel-level labelling by radiologists, to a written protocol agreed with you before work starts.
- 3D segmentation and contours
- Bounding boxes and landmarks
- Classification and structured findings
Ground-truth labelling
Reference-standard datasets built from multiple independent reads and documented adjudication.
- Multi-reader design
- Senior-reader adjudication
- Inter-reader agreement reported
Model evaluation
Independent testing of your model against held-out, radiologist-confirmed cases, with failure analysis a clinician can read.
- Sensitivity, specificity, Dice, AUC
- Reader studies
- Error review by case
How we work
From protocol to signed-off dataset.
Each engagement starts with a small paid pilot, so you can check our quality before committing to volume.
1. Protocol
We agree label definitions, edge cases and acceptance criteria with your clinical and ML leads.
2. Pilot
A small batch is read and returned with agreement metrics, so you can judge the output.
3. Production
Radiologists read at volume, with blinded double reads and adjudication where the protocol calls for it.
4. QA & delivery
Audited labels delivered in your format, with a report on agreement and disputed cases.
Scope & data handling
Built for clinical data.
We focus on cross-sectional imaging, where radiologist judgement matters most.
- De-identified data only. We work on anonymised DICOM and never ask for patient identifiers.
- Your platform or ours. We can annotate inside your labelling tool or set up a dedicated environment.
- Formats you already use. DICOM-SEG, NIfTI, RTSTRUCT, COCO and JSON.
- Full audit trail. Who read each case, when, and how disagreements were resolved.
Get in touch
Tell us what you're building.
Share the modality, anatomy and rough case volume. We'll come back with a protocol outline and a pilot quote.
hello@annotomy.com