How Verityn reads.
Verityn is a computer vision model that runs a second read on X-rays and CT scans. It flags candidate findings, draws where on the image, calibrates a confidence value, and gives a reason in plain radiology terms. Every decision stays with the clinician.
Three steps, alongside what you already do.
Nothing in the radiologist's existing process needs to change. Verityn sits alongside it.
The scan arrives
A study is uploaded through the console or streamed from PACS over a standard DICOM connection. Nothing about the radiologist's existing workflow changes.
The vision model reads it
A computer vision model reviews every region for detection and localisation, then calibrates a confidence value for each candidate finding.
Flags reach the radiologist
Findings surface in the worklist with a confidence heatmap and a plain-language reason. The radiologist reviews, agrees or dismisses, and decides.
Detect. Explain. Prioritise.
A second read is only useful if it can show where, say why, and help manage the list. Verityn does all three.
Detection and localisation
Verityn does not only say a finding may be present. It draws a bounding region on the image, so the radiologist knows precisely where to look. On CT volumes, localisation is given per slice with an indication of extent.
The model has been trained on de-identified retrospective datasets and validated against held-out reads. Its task is detection in the classical sense: flag with a location, not a report.
What comes with every read.
Detection and localisation
Verityn does not just say something is present. It draws where, with a bounding region on the image across common modalities.
Confidence heatmaps
A calibrated heat overlay shows what drew the model's attention, so a reader can weigh a flag rather than take it on faith.
A reason for every flag
Each flag carries a short, plain-language reason in radiology terms. No black box, no unexplained alerts.
Worklist prioritisation
Studies with high-confidence critical findings can be moved up the list, so the urgent work is seen first.
PACS-friendly integration
Standards-based DICOM and worklist connectors sit alongside your PACS. Results return as a secondary capture or structured report.
A complete audit trail
Every study, flag, confidence value, and human decision is logged and exportable, for governance and post-market surveillance.
What Verityn is. What it is not.
Clinical AI that is honest about its role is more useful, and safer, than one that overstates it.
- Clinical decision support, intended to assist a qualified radiologist.
- A tool that flags candidate findings and provides a reason. The radiologist reviews every flag.
- Calibrated in confidence values. A high-confidence flag is designed to mean something, and is subject to ongoing validation.
- Subject to a quality management system, version control, and post-market surveillance.
- A second reader. Not a first. Not a replacement.
- A diagnosis. Verityn does not report. The radiologist reports.
- Autonomous. There is no pathway in which Verityn acts without a clinician reviewing.
- A replacement for the radiologist. It is a tool in the radiologist's hands.
- Infallible. It misses findings and it raises false positives. Those are described in our validation documentation.
- A gateway to clinical decisions. Flags are advisory. The clinician always decides what goes in the report.
The clinician is not optional.
Verityn is designed around a single principle: every flagged finding passes through a trained clinician before any action is taken. The model does not close the loop. The model opens a question, and the radiologist answers it.
Dismissing a flag is a deliberate action, logged with a timestamp and the clinician's identifier. Agreeing with a flag is logged in the same way. Neither adds words to the report: the radiologist writes the report, as they always did.
This matters for regulatory reasons, for clinical safety, and for the trust of the radiologists who use the tool. A second reader that behaved as an autonomous agent would not be a second reader at all.
Every study. Every flag. Every decision.
A complete record from the moment a study arrives to the moment a clinician acts, for governance, post-market surveillance, and clinical safety.
Study log
Arrival timestamp, modality, study identifier, and processing version recorded for every study.
Flag log
Each flag, with its confidence value, localisation, and the model version that produced it.
Decision log
Every agree or dismiss action, with the clinician identifier and timestamp, is immutably recorded.
Model version control
Every read is linked to the exact model version in use, so post-market review can be conducted per version.
Structured export
Audit records are exportable in structured format for governance review and regulatory reporting.
Post-market surveillance
Aggregate performance metrics, flag rates, and decision patterns feed the ongoing post-market programme.
See the second read in action.
Start free on the Pilot plan and run Verityn on a small volume of your own studies. No commitment, no rip-and-replace.