Skip to content
Reza Motaghi
Lab Notes2 min read

"Normal" is not a verdict

By Reza Motaghi

"Normal" is not a finding. It is a verdict, and a radiograph cannot deliver it.

I write reports on medical images every day, and I wrote the label contract for the model I train on my own reads. One rule sits at the top of both. A read is about this image, never about the patient's health. "Not detectable on this image" is a sentence I can defend. "Healthy" is not.

A panoramic radiograph is a shadow of a curved slice through the jaws, drawn by a beam that moves. What lies outside the slice blurs or vanishes. The edge of a cortex depends on where the beam met it. Some structures are not on the image at all, whatever a report says about them. So the honest read has two parts: what the image shows, and what this image cannot rule out. The second part is where a specialist earns the signature.

Two habits I hold the model to, because I hold myself to them.

Commit. A competent read names definite findings and reserves "this image cannot show" for the few structures where it is true, never as a general hedge. Hedging everything is not caution. It is a report that says nothing and cannot be wrong.

Never end a finding with "otherwise normal". The phrase launders a real finding into a clean report, and a model learns the laundering faster than the finding.

"Normal" is the cheapest word a model can learn, because normal needs no seeing. An AI read that can only say normal or abnormal is not a read. It has no room for the line that says what this image cannot show, and that line is the part a clinician is for.

The check

Three lines that make a negative read honest:

  1. "Not detectable on this image." Never "absent". The image is the limit of the claim.
  2. Name the structures this projection cannot show, once, in a fixed place in the report, not scattered as hedges.
  3. State the finding, then the negatives, separately. Never close a finding with "otherwise normal".

About the author

DDS, MSc · Board-certified specialist in oral and maxillofacial radiology · Independent researcher

0009-0005-3194-1879

Reza Motaghi is a board-certified specialist in oral and maxillofacial radiology who reads medical images every day, 100,000+ reports signed, and runs enterprise AI in production as Chief Innovation Officer. He builds the training data from his own reads, made together with fellow radiologists under one written standard, trains models on it, and tests them on cases they have never seen.

He published Radigrade, the registered benchmark for AI-written radiograph reports: invented findings counted first, graded by an AI judge calibrated against his own verdicts, general models in public and vendors’ systems privately on request. He also built CBCTScope, the first CBCT reading viewer with native AI-agent control.

Research recordAbout RezaContactLinkedIn (opens in a new tab)

The newsletter

Get the essays as they’re written.

Imaging AI, evaluation, and building alone, from a radiologist who reads every day and tests before he trusts. A few essays a month, nothing else.

Join for FreeFree · Unsubscribe anytime

Related

  1. Know the resolution of the instrument you trust

    A vision model attends in tiles of about three millimetres on a panoramic radiograph. Most of what a radiologist calls is smaller. The one-minute arithmetic.

  2. Inventory before interpretation

    Blind on my own radiographs, the strongest general model reported a root canal in a tooth that was not there, then kept writing. Count before you interpret.

  3. What the best general model read on a specialist's ruler

    The strongest general model read three findings in ten on my cases and invented three in four. One accuracy number hides that. Three questions to ask.