Built by a radiologist who reads with it every shift
MyRadAgent is not a demo built by engineers who have never signed a report. It was built, and is used daily, by the radiologist whose signed reports supply every accuracy figure on this site.
Michael Mousa, MD
Founder · practicing diagnostic radiologist
Michael reads a full clinical schedule across CT, MRI, ultrasound, radiography and mammography, and has signed thousands of reports drafted with MyRadAgent in his own PowerScribe workflow. He built the first version for himself: a way to dictate a scratch read and get back a structured, guideline-checked draft that he would actually sign, instead of a template he would have to rewrite.
Every rule in the product, from how an enteric tube is called on a chest film to which guideline a nodule triggers, exists because it was wrong on a real case and got fixed the same day. The validation page publishes the resulting edit rates from his signed reports, with the method and exclusions.
Why it exists
Radiologists lose a measurable slice of every shift to the mechanics of reporting: turning a read into structured prose, remembering the follow-up interval for a 7 mm nodule versus a 9 mm one, keeping the impression consistent with the findings, and documenting technique and indication in the words the coder needs. None of that is the diagnostic work. All of it is where the time goes and where the avoidable errors live.
The large language models of 2026 are good enough to draft that prose from a scratch read, if they are routed, checked and constrained by someone who knows what a sign-ready report looks like. MyRadAgent is that layer: it sends each study to the best model for its lane, runs the guideline and consistency checks a careful radiologist runs in their head, learns the signing radiologist's phrasing from what they actually sign, and stays out of the way when the draft is clean.
What the product is held to
Every figure on this site says where it comes from. Edit rates come from drafts that were then signed in the app; savings estimates name their benchmark and their assumptions.
Every output is a draft for review by a licensed radiologist. MyRadAgent never signs, sends or files a report, and never renders a diagnosis on its own.
PHI is redacted before any model call, with executed BAAs behind every AI and hosting provider. Details on the security page.
A clean draft gets zero notices. A line or two appears only when the radiologist has to act, such as a conflict between findings and impression.
Models from four AI labs sit behind one router. When a better model ships, the route changes server-side; your style rules and history come with you.
A wrong call reported from a shift is traced to its cause and corrected in the product, not filed. Rejected drafts are logged so the errors radiologists actually see are the ones that get fixed first.
Where it is going
Individual radiologists first, then groups. The same product scales from one reader on a laptop to a teleradiology group with pooled volume, a shared rulebook and an administrator's view of catches and hours saved. If you read for a group and want to run a pilot, the fastest route is a short call with Michael. The group evaluation checklist describes how to run that pilot so it actually answers the question.
Contact
Questions, bug reports, a wrong call on a case, or a group pilot: support@myradagent.ai. Michael reads every message.