What iPriorAI is
A decision-support and evidence tool that answers, with published evidence and no patient data, the questions Northern Ireland's imaging AI programme currently has no tool for — starting with the one nobody asks: can you use the data on the machines?
Two questions sit underneath every imaging AI deployment
Upstream: what was the model trained on?
Was that lawful for commercial use? The most widely used open CT collections carrying expert radiologist input are restricted to non-commercial research. A supplier warranty that it holds “all necessary rights” is a contractual promise, not evidence. Can the supplier evidence its training-data provenance and commercial licence position on request — and what happens to HSC's service if it cannot?
Downstream: who owns what HSC creates?
When HSC hosts a model, contributes images to validation, has its consultants annotate cases, or generates real-world performance evidence on NI patients, value is created: tuned weights, annotations, local validation evidence, publication rights. On current market practice that value accrues to the supplier by default rather than by decision.
What iPriorAI does not do
- We are not clinicians and give no clinical advice. Clinical judgement belongs to a named clinical sponsor and to reporting clinicians involved from the start.
- We do not build machine-learning models. Certified products exist; the question is what HSC owns and is permitted to do.
- We do not work with person-level images or records outside a lawful access route. Nothing here requires access to any patient data at all.
- We do not create synthetic clinical data.
- We are independent of imaging AI vendors: no reseller, referral or revenue-share arrangement with any supplier of imaging AI.
The modules
Dataset Register — open CT collections, expert input, verified licence terms. Rights Register — the rights position per vendor class and AI artefact class, with the gap list and model contract clauses. Queue Simulator — what reprioritising the reporting queue would do, benefit and cost shown together. Evidence Library — every number traced to Published, Derived or Assumption. Governance & Regulation — the NI route, every step open. Export — a dated brief carrying the boundaries and the controlled AI System definition.
Dataset Register
A curated, versioned register of open CT collections with the expert radiologist input each carries and the licence terms that govern it. Licence terms are read at source and quoted — never inferred from a secondary description. Unverified records are excluded from every export.
Can you use the data on the machines?
The AI+IP Rights Register: the chain of rights for data and AI artefacts generated on trust-owned and vendor-operated imaging equipment and software — scanner OEM, PACS vendor, third-party AI, cloud inference, and locally derived artefacts.
Queue Simulator
A deterministic, seeded simulation of time-to-report under competing prioritisation policies. Prioritisation is zero-sum on total waiting: the benefit to prioritised cases and the cost to displaced routine cases are shown with equal prominence, on the same screen. There is no “waiting list reduction” output anywhere in this tool — attempting to display one is a defect.
Honesty requirement. Every run shows: median and 90th-centile wait for flagged true positives; the same for routine cases displaced, with equal prominence; the false-positive attention burden per reporting session; and the false negatives — cases with the finding the model does not flag — and where they sit as a result.
Evidence Library
Every number shown anywhere in this tool traces back to an evidence entry, or is visibly flagged as an assumption. No unattributed numbers anywhere.
Governance Route & Regulation
The structured chain that must be complete before any work touches a real scan. Its default state is every step open: a statement of what is missing, not a claim of readiness.
Export a dated brief
One click produces a dated brief: the register view, the latest simulation run with its parameters and sources, the evidence classifications, the rights register, the governance chain status — and, in every export, the boundaries table, the controlled AI System definition in full, and the statement that no patient data was used.
Exports print cleanly (Ctrl/Cmd+P → Save as PDF). The HTML brief carries print styling.