Use case · builds on Medical Imaging Annotation
Radiology annotation by radiologists.
Radiologists matched to the modality and body area label findings in X-ray, CT, MRI and ultrasound studies and their reports. Independent reads, senior adjudication and reader agreement analysis are matched to how hard the task is, giving you labels and reference test sets you can defend.
- Who it's for
- Radiology and medical imaging AI teams.
- The problem
- Radiology AI depends on reads that are consistent, documented and right for the modality.
- Clinical and data science expertise
- Radiologists with the right modality and subspecialty experience, with PhD data scientists who analyse reader agreement.
- How quality is assessed
- Independent double reads where needed, senior adjudication and reader agreement analysis.
NDA and data processing agreement on every project. How we handle your data.
Images, reports and model outputs.
Classification and grading
Assigning findings, grades or categories to images or studies.
Localisation
Marking where findings are, using the annotation method your project needs.
Report labelling
Extracting and labelling findings from free-text radiology reports.
Review of existing labels
Checking labels you already have for clinical accuracy and consistency.
Reference test sets
Radiologist-reviewed cases for testing how your model performs.
AI output review
Radiologists assess your model's outputs or draft reports and explain the errors.
- Radiologist-reviewed labels
- Annotated datasets
- Reference test sets
- Quality assurance findings
- Reader agreement analysis where appropriate
- Documented annotation criteria
Working in ophthalmology, pathology, dermatology or endoscopy? See medical imaging annotation.
Matched to how hard the read really is.
- 01
Single expert read
One radiologist per study, with quality checks on a sample. Suits clear-cut tasks and early prototypes.
- 02
Independent double read
Two radiologists read each study without seeing each other's work. Where they disagree, a senior reader adjudicates.
- 03
Agreement analysis
PhD data scientists measure how often readers agree, so you know how reliable the labels are and where the task is genuinely hard.
Decisions we settle with you.
- Which modality, body area and subspecialty experience does the task need?
- Is one read enough, or do you need independent reads and adjudication?
- What is the reference standard: reader consensus, or a confirmed outcome where one exists?
- How should incidental findings, borderline and non-diagnostic studies be handled?
- What annotation method and output format does your pipeline need?
Configured per project.
Illustrative example
A study is read independently by two radiologists, with senior adjudication where they disagree. Where they agree, the read becomes the reference label.
- Reader A · independent read
- Reader B · independent read
- Disagreement highlighted
- Senior adjudication
- Reference label locked
Radiology annotation, answered.
Anything else, just ask on a scoping call.
Who annotates the images?
Radiologists matched to the modality and body area of your project, drawn from the BiteLabs network of 30,000+ UK and US clinicians. DeepLabel is a sister company to BiteLabs. We check professional registration and experience, and confirm the right expertise for your project during scoping.
Which modalities can you cover?
X-ray, CT, MRI and ultrasound. We confirm the modalities, body areas and level of subspecialty experience your task needs during scoping.
Can you label radiology reports as well as images?
Yes. Radiologists can label findings in free-text reports, check report-derived labels against the images, or review AI-generated report text for accuracy and omissions.
How do you make labels reliable?
We match the review design to how hard the task is. That can be a single expert read with quality checks, or independent double reads with senior adjudication where readers disagree. PhD data scientists measure agreement between readers so you know how reliable the labels are.
Can you work with our existing labels and tools?
Yes. We can review labels you already have for clinical accuracy and consistency, and agree the annotation method and output format your pipeline needs during scoping.
How do you handle imaging data?
We offer an NDA and put a data processing agreement in place on every project. Every reviewer signs a confidentiality and data protection agreement.
Related services.
Let's solve your next clinical AI challenge.
Tell us what you're developing and we'll reply the same working day.