Radiologist | AI Entrepreneur @DocSmart_AI | Guitarist ๐ธ | MMA Enthusiast ๐ฅ | Building the future of AI in radiology while staying sharp in & out of the ring
We fine-tuned DKFZ's nnFoundation on our lymph-node CT: about a quarter fewer false positives than our from-scratch UNet at similar recall, on 27 scans. Early results, no held-out test yet.
Paper: https://t.co/tHZtz1W3JU
Code: https://t.co/TfQMwzprZ9
#Radiology#MedicalImaging
All who praise a model like opus 5.5 with memes and then later complain, remind me of children on a roller coaster ride.. it's very obvious the AI companies are playing everyone by tweaking same models, temporarily boosting them pre release and then nerfing them b4 next release.
LungIMPACT (Nature Medicine): AI worklist prioritisation of 93,000 GP chest X-rays in five NHS trusts. Reports came back a median 13 hours sooner. Time to lung cancer diagnosis was 44 v 46 days (P=0.84). The authors advise against it here.
https://t.co/eM9YCcTWzG
#Radiology
@rohit3a Even Fable not at par with Sol 5.6 let alone Astra 6.. it's defo learnt better English ill give it that but still useless in complex coding tasks afaik
@AtlasOmeg@Mr_Salio One day ppl will lean about hypothesis testing :P i say there's mounting evidence that they nerfed it. And tui might say it's only the bad reports reaching us.. then we debate endlessly ..
@x86machine @thesupermanmx Exactly what claude fact check told me. Second day in a row I've seen a liar as a newbie to x.. true to the nature of the owner. Coming from LinkedIn, x is like a busy smelly fishmarket,
Geoff Hinton's 2016 call โ radiologists obsolete within five years โ aged like milk; in 2026 demand is the strongest ever. But the real threat was never AI itself. It's other specialists using AI. My full argument: https://t.co/YEvzHktbFx #Radiology#AIinHealthcare