RadiologistRaj_MD OP
Senior Radiologist · Tier-2 hospital · 14 yrs
Been using a CAD (Computer-Aided Detection) tool for chest CT scans for about 10 months now. Honestly its picking up nodules I would've flagged anyway, but occasionally it catches something I probly would have scrolled past at 11pm after 40 scans. That's real. The vendor says AI diagnostics improves sensitivity by 30% — I believe maybe 12-15% in real conditions. Anyone in a NHS or AIIMS setup using these? Curious if the results hold in high-volume public hospital settings where scan quality isnt always ideal.
NurseParamita_ICU
ICU Nurse · Public hospital · Mumbai
Not radiology but relevant — we use an AI early warning system (EWS) in the ICU that flags sepsis risk before the patient deteriorates. False positive rate is annoyingly high (~40%) but the true positives it catches are genuinely life-saving. We've had 3 cases in 6 months where the AI flagged sepsis 4-6 hrs before any clinician would have suspected. AI in critical care is real, not hype — just needs tuning to your patient population.
HealthIT_Lucas
Health Informatics · NHS England
NHS has been piloting AI radiology tools under the NHS AI Lab since 2021. The published data from the DART (Deploying AI Responsibly in radiology Trials) program shows about 11% improvement in early-stage cancer detection across breast screening. But — and this is huge — performance drops significantly when the AI model was trained on US data and deployed on UK patient demographics. Dataset bias in medical AI is a massive unsolved problem, not a small asterisk.
DrKumar_Onco
Oncologist · Private oncology centre
The dataset bias point is SO important and still underreported in mainstream AI in medicine coverage. Most FDA-cleared AI diagnostics tools were validated on predominantly white western cohorts. We see degraded performance on South Asian and African patient populations regularly. This isnt speculation — a 2023 paper in Nature Medicine quantified it: AI diagnostic accuracy dropped by avg 7.4% when tested on underrepresented groups. Regulators are starting to mandate diverse validation sets but its slow.
ZainabMed_Student
Final year MBBS · Interested in health tech
Reading this thread as someone about to enter the field — genuine question: should we be worried about AI replacing radiologists? I keep reading conflicting stuff. Some say radiology is most at risk, others say AI will just be a tool. Whats the actual concensus among practicing radiologists rn?
