Every scan is a stack.
Only one slice matters.
RealActivity Imaging scores every slice in a series independently, tells you which one drove the verdict, and shows you the region the model actually attended to — across six cancer pathways, in one platform, on credentials your hospital already has.
This is a research platform, not a diagnostic device. It holds no FDA clearance and no CE marking. It supports clinician assessment and does not replace clinical judgement — every output requires review by a qualified clinician before any diagnostic or treatment decision. The scan above is synthetic, not a patient.
The same shape of answer, every time
A verdict is not enough to act on. Every analysis returns the codes a clinician needs for the chart, the confidence behind the call, and a visual account of what the model responded to — in one structured object that exports as FHIR R4.
Six pathways, at honestly different stages
Every model is benchmarked against public datasets before it is deployed. Independent clinical validation is in progress per pathway, starting with spine. We publish where each one stands rather than averaging them into a single number.
Skin
DermoscopyThe most extensively benchmarked pathway, on the public ISIC archive, including a published skin-tone fairness analysis across the Fitzpatrick scale — sensitivity and false-positive rate stratified by skin type, with real Fitzpatrick IV–VI sample sizes (n=32, n=5) too small yet to draw a fairness conclusion either way.
BenchmarkedBrain
MRITumour type and grade classification on public MRI datasets, with occlusion-based localisation rather than a second segmentation model.
BenchmarkedLung
CTTwo-stage nodule detection: candidate generation across the series, then classification. Benchmarked on public thoracic CT.
BenchmarkedBreast
MammographyCC and MLO view classification. Re-validated on an external cohort after a domain-balanced retrain; external specificity remains materially below internal.
External gapSpine
MRILumbar degenerative severity grading. First pathway scheduled for a prospective clinical validation study.
Validation nextLiver
CTSegmentation with tumour sub-detection. External Dice 0.77 for liver, 0.48 for tumour sub-detection — still unreliable, and labelled as such in the product.
Research previewThree numbers a probability alone will not give you
Accuracy on a benchmark tells a hospital very little about the scan in front of them. These travel with every result, so a clinician can calibrate how much weight to give it — case by case, not vendor-wide.
Is it looking at the lesion?
Scores how much of the model's attention falls on the actual finding rather than background or artefact. Above 0.7 the attention is on-target; below 0.5, treat the verdict with suspicion even when confidence is high.
How stable is the call?
The same scan is run under multiple augmentations. A wide spread means the result hangs on framing or noise rather than on pathology.
Which slice decided it?
Every slice in the series is scored independently, so you see whether concern is one focal peak or smeared across the volume — which usually means artefact, motion, or multifocal disease.
Built for the people who have to sign it off
A result is only useful to a clinician who can see why it was produced, and only adoptable by an institution where every action is accountable afterwards.
Audit log on every action
Sign-ins, scans, clinician feedback and admin actions are recorded with user, timestamp and detail, reviewable from the admin panel.
FHIR R4 and ICD-10 out of the box
Results export as FHIR R4 DiagnosticReport resources carrying ICD-10 codes. SMART on FHIR launch is implemented and tested against a public sandbox.
Your existing Microsoft identity
Clinicians sign in with hospital Entra ID credentials — no new account, no separate password store. Four roles scope what each account can see and do.
Hospital-scoped by design
Every scan, report and admin view is scoped to the institution that produced it, and clinician overrides feed an active-learning queue rather than disappearing.
What runs today, and what is still being built
Stated plainly rather than presented as three equally available options.
Cloud-hosted
Register your institution and clinicians sign in with Microsoft the day approval completes.
- No infrastructure for your IT team to run
- Hospital-scoped data isolation
- Managed model updates
On-premises
Full data sovereignty. Infrastructure-as-code exists; it has not yet been provisioned or load-tested against a live environment.
- Scans never leave your network
- Your own identity and audit boundary
Integrated enterprise
Direct EHR and PACS integration. SMART on FHIR launch is real and sandbox-tested; a production EHR integration has not yet been run.
- FHIR Subscription push to your systems
- Results into the existing worklist
What registration actually involves
Institutional registration is reviewed by our clinical team, typically within two business days.
Apply
Institutional details, the pathways you need, and your intended use case.
Review
Our clinical team checks the application against the platform's current scope.
Configure
Entra ID tenant setup and role assignment for your clinicians.
Go live
Your hospital is provisioned and clinicians sign in with existing credentials.
Register your hospital
Or read the methodology first — model cards, validation status and API reference.