The problem: Emergency radiology can’t recruit its way out of this…
Ireland has around 6.5 radiologists per 100,000 people, a little over half the EU median. Meanwhile, out-of-hours CT demand keeps rising. Medica already provides overnight acute CT reporting for 20 of Ireland’s 28 emergency departments.
And yet most of what comes through the door is normal: around 9 in 10 emergency CT brain studies, and roughly 3 in 4 MSK trauma X-rays, carry no acute finding. That normal majority still consumes the same scarce reporting time as everything else, and it’s the one part of the workload demonstrably possible to handle differently.
The idea: One signal, two uses
High-confidence normal triage uses AI to predict which studies are normal, tuned for very high negative predictive value, because in triage, the cost of a missed positive is what matters, not overall accuracy.
That single signal does two distinct jobs:
Prioritisation — surfaces the critical minority faster. Nothing is ruled out; the order changes, so suspected brain bleeds reach a radiologist sooner.
Capacity release — lets the confirmed-normal pool be deferred or ruled out under governance, freeing radiologists to focus on complex work and speeding emergency department flow.
Both keep the radiologist reading every study and accountable for every report.
The evidence: Two live deployments. Real clinical service, not a trial
qER (Qure.ai) – CT Brain
- ~200,000 studies over 14 months
- 90% predicted routine
- ~0.98 negative predictive value vs continuous expert audit
- ~19K suspected brain bleeds surfaced and expedited
RBfracture (radiobotics) – MSK trauma
- ~70,000 studies over 6 months
- 78% predicted no fracture
- 0.97 negative predictive value at pre-live validation
- 1,800-2,700 radiologist reporting-hours identified in the normal cohort
Governed by design: Built to be trusted, not just to work
Every deployment runs as a radiologist-plus-AI model, the human reads, and remains accountable. Performance is continuously audited against expert re-read. Operating points are clinically owned, not vendor-set. Risks like automation bias and performance drift are actively monitored, not assumed away.
The approach is built around the EU AI Act’s human-oversight requirements for high-risk clinical AI, HIQA’s national AI guidance, and Ireland’s AI for Care 2026–2030 Strategy.
What Medica proposes: A repeatable model, not a one-off tool
Medica is proposing a governed Irish pathway for evaluation and adoption: defining appropriate emergency imaging use cases, validating performance locally, agreeing the operating point, retaining radiologist accountability, and monitoring continuously for safety, drift and measurable benefit.
Not simply a deployment, a repeatable model for responsible AI adoption in emergency radiology: gated for autonomy, scoped by population, grounded in local validation, and built to keep radiological judgement central.
To read the full whitepaper:
Get the complete evidence, methodology, governance framework, and Medica’s phased roadmap for responsible AI-as-a-service in emergency radiology, by filling in the form.
For all other enquiries, please contact: info@medica.ie