A second reader, never the signer

Bangladesh has roughly seven hundred radiologists for more than one hundred seventy million people, about four per million, almost all of them in Dhaka. A film taken in a district town can wait days for a read. The obvious thing for a capable vision model to do is read the film itself and skip the wait. It is also the wrong thing.

Glyph's role at a diagnostic center is not radiologist replacement. It is co-interpretation. The model offers a draft observation, a likely finding, an urgency flag, to the technologist who took the film, and that draft routes to a remote radiologist who verifies and signs. The model reads first; the licensed human reads last and signs. The scarce specialist's hour is spent confirming structured drafts from many centers instead of reading cold films one at a time for one.

The difference between replacement and co-interpretation is the difference between a report no one will trust and one that travels. A film read only by a machine is an unsigned guess, and the country has already learned what happens to diagnostic reports no accountable human stands behind. A film drafted by the model and signed by a radiologist is the radiologist's read, made faster. The scarcity is eased without the signature being cheapened, because a licensed human still puts his name on every report.

This is where the model belongs in imaging: as the second reader that makes the first reader reachable, not as a substitute for the reader the country does not have enough of. It multiplies the four-per-million instead of pretending to be the four-million-and-first. The specialist stays in the loop, because his signature is the thing that makes the read worth anything, and the model's gift is to spend his scarce attention only where it counts.