Image analysis
Segmentation and detection across MRI, CT, X-ray and digital pathology. Every finding traces back to the voxels that produced it.
4 modalitiesDICOM native
Clinical AI for imaging and risk
Know which twelve studies to open first.
Bedside pre-reads every study and every new observation, ranks what needs a clinician now, and shows the evidence behind each flag. A clinician signs every report.
Pre-read · MR AX T2 FLAIR
Model v4.2 · 1.8 s
Focal FLAIR hyperintensityRight frontal · 12.4 mm · prior from 2025-11-04 queued
Confidence0.93
In production at
Northvale Health
Meridian Clinics
Halden Institute
Cortis Radiology
Provia Care Group
Aldermoor Trust
Outcomes · 18-month rollout, six sites
Signal you can act on, in the time you actually have.
Each figure below is footnoted with its method and links to an auditable model card.
- 141,800 studies across six sites, Jan 2025 to Jun 2026. Consensus of three board-certified radiologists, blinded to model output.
- 2Median change in time-to-first-read for studies flagged urgent, compared with the 12 months before deployment.
- 330-day all-cause readmissions in high-risk cohorts, against a matched pre-deployment period.
- 4Two academic centres, three outpatient clinics, one national trust. All figures on this page are synthetic.
Platform
One platform, four clinical workloads.
Every model ships with its own evaluation set, a drift monitor and a plain-language explanation for the clinician reading it.
Predictive risk
Deterioration, sepsis and readmission risk, recalculated on every new observation instead of once a shift.
6 h lead timeHL7 v2 · FHIR streaming
Treatment planning
Options ranked against the patient's own history, comorbidities and genomic signals, with the trade-offs shown.
Guideline-awareExplainable ranking
Research cohorts
Build a de-identified cohort from a plain-language query and export it to your notebook with provenance intact.
De-identifiedOne-click export
Predictive risk
Risk that moves before the patient does.
Scores recompute on every observation. Pick a cohort to see the last 12 hours and where the alert threshold sits.
Sepsis risk · Bay 4
Updated just now · recomputes on every observation
Current score
0.81
+0.07 in the last hour
Alert threshold
0.60
Above for 2 h
Risk score
Uncertainty
Alert threshold
Sepsis · 6 h median lead time before the first vital sign crosses its limit.
-
Bay 4 · PT-4021958F · post-op day 2Escalate0.81
Rising lactate, HR +18 over 4 h, falling urine output. Escalate per sepsis pathway.
-
Bay 9 · PT-3882074M · CHFReview0.46
Weight up 1.8 kg in 3 days with SpO₂ drift. Review diuretic dose.
-
Bay 12 · PT-4100733F · day 1Stable0.09
Stable across all tracked signals. Next recompute in 15 min.
How a pilot runs
It lives inside the workflow you already have.
No new window to learn. Results land in the PACS worklist and the chart your team already opens.
Weeks 1-2
Connect
Point your PACS and HL7 or FHIR feed at an endpoint in your own VPC. Data stays in your region, and nothing trains on it without a signed addendum.
Weeks 3-6
Calibrate
Run in shadow mode. Compare model output with your own reads, then set alert thresholds for each service line.
Week 7 onward
Go live
Findings arrive as a pre-read with the evidence attached. Drift monitors and a weekly model card keep your governance board current.
“The value wasn’t replacing a read. It was knowing which twelve studies to open first.”
Governance
Built for the people who have to sign it off.
Everything a clinical safety officer asks for on day one, available before you talk to sales.
Regulatory status
Listed per model, with validated populations
Cleared indications, listed in full
- FDA
- 510(k) number placeholder
- EU / UK
- CE · UKCA class IIa placeholder
- Intended use
- Triage and decision support. Not for standalone diagnosis.
- HIPAA and GDPRBAA and DPA on request
- SOC 2 Type IIAudited annually
- Model cardsTraining data, limits, drift
- Full audit trailEvery score and override
- Version rollbackFrom the console, no redeploy
Questions
The ones your board will ask.
Anything not covered here goes to our clinical safety lead, who replies within one working day.
Ask a questionNo. It produces a pre-read with confidence scores and the evidence behind them. A clinician reviews and signs every report, and every override is logged and added to the evaluation set.
In your region, in a VPC you control. De-identification happens before anything leaves that perimeter, and no customer data is used for training without a signed addendum naming the exact cohort.
Two weeks to a shadow-mode deployment for a typical PACS and HL7 feed. Then four weeks of calibration against your own reads before anything reaches a clinician.
Misses and false positives are reviewed weekly with your clinical safety officer. Thresholds can be tuned per service line, and any model version can be rolled back from the console.
Drift monitors compare live inputs and outputs with the validation set every week. A drop past the agreed limit alerts your safety officer and pauses that model's flags until it is reviewed.
Clinical pilot · booking for Q1 2027
Bring AI into the read, not around it.
Start with one service line and a four-week shadow deployment. If the numbers don’t hold up against your own reads, you walk away.
- Nothing reaches a clinician until shadow mode is signed off
- Patient data stays in your VPC and your region
- No exit fee if you stop at the week 6 decision