Here is a thing that sounds like it should be illegal and is, in fact, blessed by the U.S. government. You are a diabetic. You go to your regular doctor - not an eye specialist, your regular doctor, the one with the blood-pressure cuff and the sad little jar of tongue depressors. A technician points a camera at the back of your eye. Thirty seconds later a piece of software says, on its own authority, "this person has diabetic retinopathy" or "this person does not." No ophthalmologist looks at the image. No human signs off. The machine diagnosed you, and that is the point, and the FDA said that was fine.
That is the whole business of Digital Diagnostics, a company in Coralville, Iowa, which is not a place you associate with regulatory firsts in artificial intelligence, which is part of why this is fun. In 2018 its product - then called IDx-DR, now called LumineticsCore - became the first autonomous AI system the FDA ever cleared to make a diagnosis without a physician reviewing the result. Not "assist." Not "flag for review." Diagnose.
The word that does all the work is "autonomous"
Most companies that say "AI in healthcare" are quietly selling you a very fancy highlighter. The software looks at a scan, circles the scary parts, and hands it to a doctor who makes the actual call and takes the actual responsibility. This is a comfortable arrangement for everyone, especially lawyers.
Digital Diagnostics does the uncomfortable version. LumineticsCore makes the clinical decision itself, at the point of care, and the company took on the medical liability for that decision. If you are going to let a machine act like a doctor, someone has to be liable like a doctor, and the founders decided it should be them. That is a business model and also, if you squint, a philosophical position.
The founder is his own least likely hire
Digital Diagnostics was founded in 2010 by Michael Abramoff, who is a fellowship-trained retina surgeon, a neuroscientist, and a computer engineer, which is three careers that normally do not appear on one resume. This matters, because the hard part of autonomous medical AI is not the neural network. The hard part is convincing a regulator that the network is reasoning about the disease the way a specialist would - not just pattern-matching pixels in a way that happens to work until, one day, it spectacularly does not.
So Abramoff built the thing around biomarkers - specific, nameable clinical features - rather than an inscrutable black box. The AI can, roughly, show its work. This is the sort of decision that is invisible in a demo and load-bearing in an FDA submission. It is why the company likes to say it does "AI the right way," and, tellingly, why its Facebook handle is literally "AItheRightWay." In a field where "the right way" is usually a slide, here it is a URL.
Why this exists: the exam nobody gets
The clinical problem is depressingly simple. Diabetics are supposed to get a retinal exam every year, because diabetic retinopathy is a leading cause of preventable blindness and it is very treatable if you catch it early. Enormous numbers of people skip the exam - it means another appointment, another specialist, another day off work. The disease progresses in the gap.
Digital Diagnostics' insight was not a better algorithm so much as a better location. Put the exam where the patient already is - the primary care office - and let software do the reading so you do not need a specialist in the room. The specialist's time is the scarce resource; the software is not. When the machine says yes, you get referred to a real eye doctor for treatment. When it says no, you go home, exam done, care gap closed.
The money followed the milestone
The company has raised more than $130 million, including a $75 million Series B in 2022 led by KKR, with 8VC, Optum Ventures, OSF Ventures and others along for the ride. Investors did not back a promise; they backed an authorization that is genuinely hard to copy. In most software, the moat is the product. In regulated medical AI, the moat is the regulator - the years and the evidence it takes to get through the FDA - and Digital Diagnostics got through first.
That head start is now being widened the unglamorous way: distribution. LumineticsCore runs on a Topcon retinal camera; a partnership with Baxter aims to push autonomous screening into more care settings; health systems like OSF are deploying it to actual patients. None of this is a viral consumer app. All of it is the slow, real work of getting a diagnosis into a room where it wasn't before.
What you can actually do with it
If you run a primary care practice or a community clinic, the pitch is concrete: add diabetic eye screening you can bill for, without hiring an ophthalmologist or shipping images out for a read. If you are a health system, it is a way to close a stubborn care gap across a lot of sites at once. And if you are a patient, it is the difference between a screening that takes thirty seconds during a visit you already scheduled and one you were never going to make.
There is a version of AI in healthcare that mostly widens the gap between well-resourced hospitals and everyone else. Digital Diagnostics aimed the other way - at the clinics where the specialists aren't. Whether autonomous diagnosis scales to the many diseases the company would like it to is an open question. But the first one already works, in real clinics, with the FDA's name on it. Which, again, sounds like it should be illegal, and isn't, and that's the interesting part.