While everyone watched the model releases this week, the FDA quietly cleared something more consequential for actual practice: a PACS imaging viewer with AI built directly into the platform, not bolted on as a separate tool.
That distinction is everything. Most clinical AI today lives in a different tab. It works, and nobody uses it, because "open another system, log in, copy the result back" is friction that a busy radiologist or oncologist will route around every single time.
AI that lives inside the tool you already have open is a categorically different adoption story. No new login. No new workflow. It's just there when you're already looking at the scan.
This is the unsexy truth about clinical AI: the winner in most specialties won't be the best algorithm. It'll be the one that got embedded where the work already happens. Distribution beats intelligence in medicine, almost every time.
The headlines will keep going to the frontier models. The clearances that change my Tuesday will keep going unnoticed.
https://t.co/RH5vKQZO7g
This vial contains a new drug called PAC-832, which I recently invented to treat Alzheimer’s disease. It is the world’s first selective GalR1 antagonist.
I designed and synthesized PAC-832 in a chemistry lab I built in my garage. (1/16)
@cunha_tristan@snickersigl@SebastianCaliri The "bitter lesson" might better be taught if they didn't start with crappy images. There are public MRI and CT databases with much better imagery, a higher percentage of pathology, and accurate ground truth.
@cunha_tristan@snickersigl@SebastianCaliri Not at all confident. There's a million other med technologies I can say the same thing about. This tech stands out because of their high profile and outrageous unsubstantiated claims.
Also, there's no AI here.
@nickmmark You are right on the money. And every bit of their presentation is outrageous hype.
50,000 units worldwide!
1 billion scans a month!
30% lower mortality!
50% healthcare cost reduction!
And yet not one single image of any pathology.
@jasoncrawford Learn screening literature for renal cell carcinoma. Curable when caught early, deadly when metastasized. Can image with US, CT, or MRI. Ask your AI of choice "Explain the pros and cons for screening for renal cell carcinoma in high and normal risk patients."
@jasoncrawford It's based on studies where screening is actually performed and the results of the screening are assessed. This isn't a new idea. It's been heavily evaluated for every organ in the body using imaging far better than the USCT from Midjourney.
The Midjourney whole body ultrasound CT has some promise. USCT has been held up by technical limitations which maybe they have overcome. It could be a welcome new imaging modality. But Midjourney's cringeworthy overselling raises red flags.
@drjohnm The statistic is misleading. To frame it, imagine there were no deaths in the treatment group at all. What do you think would be the number of days of life saved?
To understand RFK Jr, realize that he sees everything through the eyes of a plaintiffs attorney. "Who can I get away with sueing?" Is the only question he asks. His "link" between Tylenol and autism with be brand name only. Expect hardly a word about acetaminophen.