A senior subspecialty radiologistreviewing every case with you.
Expert Radiology is a workstation for case review, structured report assistance, and rule-based classification across the main RADS systems — built for radiologists who want a senior second reader in the loop. Not a chatbot, and not a replacement for clinical judgement.
Cirrhotic patient. 24 mm segment VIII observation, APHE, washout on PV phase, no capsule, no prior imaging.
LI-RADS LR-5 — definitely HCC. APHE + washout at ≥20 mm meets LR-5. Recommend multidisciplinary referral per institutional pathway.
From de-identified findings to a defensible category
Describe the case
Cirrhotic patient. 24 mm segment VIII observation. Arterial phase hyperenhancement, washout on portal venous phase. No capsule, no prior imaging, no tumour in vein.
No names, no identifiers — an automated screen blocks obvious PHI before anything leaves the workstation.
The pipeline runs in the open
- Parse findings — nothing is inferred or filled in
- Retrieve evidence from a verified knowledge base
- Apply the LI-RADS v2018 rule table (every cell visible)
- Three consultant models + a structural skeptic argue it out
- Arbitration, then per-claim citation verification
Model agreement never raises confidence. Only retrieved evidence does.
Get a structured, cited answer
≥20 mm with APHE + washout meets LR-5 criteria. Suggested report wording, differential, and the exact rule cells applied — each claim tied to a citation that was actually retrieved.
Missing inputs? The helper returns "cannot assess" and lists what it needs — it never guesses.
Built around how reporting actually happens
Expert second-reader review
Submit a case and get a structured senior-consultant opinion: most likely diagnosis, reasoning, differential, suggested report wording, and a confidence rating tied to retrieved evidence.
Report review & wording
Paste a draft report and get safer, report-ready wording with important negatives, medico-legal pitfalls, and an improved impression.
Rule-based classifiers
Deterministic helpers for LI-RADS, PI-RADS, BI-RADS, Bosniak, Lung-RADS, TI-RADS, O-RADS, and RECIST. The rules are visible; missing inputs return "cannot assess" instead of a guess.
Differential & second reader
A ranked differential with why-for / why-against per item, plus an explicit second-reader review of your stated impression before sign-off.
Most AI tools ask you to trust them. This one shows its work.
Citations must exist — and must support the claim
Final answers may cite only evidence chunks actually retrieved from a verified knowledge base. A second check verifies each cited chunk actually supports the claim it is attached to. The system never invents a reference.
Agreement never upgrades confidence
Several models sharing one wrong memory is not consensus. The arbitrator is instructed that only retrieved evidence raises confidence; unsupported unanimity is capped at Moderate.
Rules you can audit, version-pinned
Classification helpers are deterministic code, not model recall. Every rule cell applied is shown, and each helper states the guideline version it implements.
An honest 'cannot conclude' state
Missing inputs, major disagreement on weak evidence, or failed retrieval trigger an explicit hard stop with reasons — instead of a confident-sounding guess.
Eight RADS helpers with visible logic
Each helper is a deterministic rule engine that constrains the category, refuses to fill in missing inputs, and shows the exact decision logic it applied.
Ali Moshibah
Consultant radiologist and builder of practical radiology AI tools. Expert Radiology exists because the most useful second opinion is the one that arrives before sign-off — structured, cited, and honest about what it doesn't know.
A second reader in the loop, starting with your next case
Closed beta, invitation only. Tell us who you are and how you'd use it — invitations go out by email when there's a good fit.
Request access