When is a medical device not a medical device? Great paper in NPJ Digital Medicine. The answer is never - but how do we regulate to ensure safe and effective use of general purpose systems? www.nature.com/articles/s41...

Now hiring - Clinical Associate As a Clinical Associate (a junior/mid-level role), you will play a key part in guiding the product strategy, clinical evaluations, and regulatory submissions for our clients developing Software and AI medical devices. www.hardianhealth.com/careers-list...

Did you catch our MD @drhughharvey.hardian.health on the Future Medicines AI podcast? He shares his expert insights into the challenges and opportunities in #healthcare #AI and explains what it takes to successfully bring AI #medicaldevices to market. Listen here ➡️ www.fmai-hub.com/videos/navig...

📋 Ready to elevate your QMS game? Part 2 of our QMS blog shares the practicalities of implementing & managing a QMS effectively. Whether you’re just starting out or optimising your existing system, this is your go-to guide for navigating the process⬇️ www.hardianhealth.com/insights/imp...

Merry QMSmas one and all 🎄

Absolute class from Gukesh, the new world chess champion. Not only the youngest player ever to achieve the title, but also a total sportsman. Calm, collected, humble - he even stays behind to clear the board like a total pro. www.youtube.com/watch?v=sbPe...

Regulations do not stifle innovation. Spread the word. Full interview: www.fmai-hub.com/videos/navig...

Excellent podcast on #AI #regulation in #radiology! Great thoughts and insights shared by @asset25.bsky.social, @drhughharvey.hardian.health!! Hosted by @sublunar.bsky.social and @auntminnie.bsky.social. m.soundcloud.com/auntminnie-c...

See you all there #RSNA24 Regulatory Landscape of AI in the US and Beyond | W1-CIN18 Wednesday, Dec 4 8:00 AM - 9:00 AM CST reg.meeting.rsna.org/flow/rsna/rs...

The FDA has lost control of clinical AI. It pains me to say so, it really does, but this is the feeling I got walking around the AI showcase here in Chicago at #RSNA24. Booth after booth, row after row, unregulated AI products and claims were everywhere....

Here's to all the #RSNA24 Europeans wide awake at 5am in Chicago waiting for the sun to rise. Good thing that tomorrow there's a breakfast event hosted by Annalise starting at 6:30am in the Great Lakes Room at the Marriot Marquis. See you there! annaliseairsna24.rsvpify.com

Who is responsible for clinical AI bias and discrimination? You, the user, are. At least, in America, according to RSNA under federal law ACA Section 1557 www.rsna.org/-/media/file...

I'm off to #RSNA2024 in Chicago. Topics I'm looking forward to: Large Vision Models - how accurate are they per pathology compared to dedicated DL trained on label data? LLMs for reporting, catching missed findings, monitoring for downstream follow up How to fit these into regulatory frameworks

Repeat after me: Any software intended for diagnosis, prevention, monitoring, prediction, prognosis, treatment or alleviation of disease is a medical device, and requires regulatory authorisation. Why do people still not know this? Let alone, get funding to build it without any regard to the law?

Can 'self-learning' AI get regulatory clearance as a medical device? Not according to the team of Notified Bodies in the EU. media.licdn.com/dms/document...

Medics on Bluesky - you can label your account with up to 4 medical specialities here bsky.app/profile/meds...

Another hint from the @fda.gov that they are considering regulating ambient scribes. The FDA also warns manufacturers against using undocumented foundation models, since training data and bias cannot be controlled. www.fda.gov/media/182871...

Thanks to @radiopaedia-org.bsky.social for hosting this session on AI regulation in radiology. www.podbean.com/media/share/...

Want to know the latest international standards for AI medical devices? I've broken them down into essential base standards, best practice guidance, and cutting edge new standards. https://www.hardianhealth.com/insights/every-regulatory-ai-medical-device-standard-you-need-to-know

If AI is 'so good at coding and will replace developers' - why are OpenAI still hiring coders?

Tell me again that medical device regulations aren't sexy. Apple's 6th FDA approval (pending) allows the airpods to be used as hearing aids, and conduct a clinical-grade hearing test.

I'm not a fan of slick marketing campaigns. It's better when happy clients do the talking.... A zero non-conformity audit is a rare thing, so well done to the @MycardiumAI team, and thanks for working with @HardianHealth

reg/acc

Interested in the cutting edge of regulatory science for AI medical devices? Want to better understand NICE EVA, or NHS procurement? Or just fancy networking with people passionate about healthtech? Sign up to get the ticket announcement on 1st...

Just your friendly reminder that #GPT4O terms of use prohibit use for medical purposes, and that medical device regulations apply for any intended purposes such as diagnosis, monitoring, treatment, alleviation or prevention of a disease or injury. Carry on.

Do EPIC's generative AI replies to patient queries actually save time? No. 17.9% INCREASED reply length 21.8% INCREASED read time -5.9% DECREASED reply time (non-significant) ..."not anywhere near where it needs to be to be...

Do you know what would be even better? Providing the required documentation for developers to build *regulated* applications off your model. Training data details, risk management processes, QA documentation etc Otherwise, you can't use it "anywhere without restrictions".

Me saying stuff to @JamesSomauroo for an hour. Have a listen. https://www.youtube.com/watch?v=ffOig7QOBcw

LLMs in medicine - what are the risks? One risk I believe we are sleep-walking into is what I call the 'exponential hallucination problem'. Let's assume we have an LLM that can ambiently scribe clinical notes from a consultation, and provide a summary with recommendations.

Peer review is broken.

The AI Emperor has few clothes.... https://link.springer.com/epdf/10.1007/s00330-024-10664-0?sharing_token=zyag2RFNONnYk-1wcFs1Gfe4RwlQNchNByi7wbcMAY6NARMmc8h3o8EM8QfDHhlJGFakpl5OnPngGceKUtiBIRhQvkU-IQAQ4At10oHFhY0E9ckRY8Y-umNbfBUKdwn3v4V_2f90MkH2Heai0H_DJ3mZOVyZqweKjzOD4e9zofI%3D

Here are my five suggestions that the UK government and NHS should do to accelerate radiology AI adoption: 1) Scrap the DTAC! 2) Change IR(ME)R 3) Model the economics at national scale 4) Develop an AI procurement framework 5) Incentivise...

"Open source is safer than closed source" I never bought this argument. And now I definitely don't... https://thehackernews.com/2024/03/over-100-malicious-aiml-models-found-on.html

Health and wellness apps don't always need a CE mark, but they should still aspire to be quality products. We've made it easy for app developers to check their compliance levels with our self-assessment tool. Unlimited usage, completely free. You're welcome!

There is a train of thought that argues for a lower regulatory burden for smaller companies developing AI. I'm not sure I agree with it... Since AI systems can be deployed across the whole economy, interacting with millions of people, they are likely to pose systemic risks, but

If your New Year's resolution is to get your healthtech: Regulatory approved ✅ Clinically evidenced 🩺 Economically modelled 💰 IP protected 🔒 ...then get in touch. 2024 is going to be a smasher.

Anyone know what problem LLMs were originally designed to solve? And if they weren't designed to solve a problem, what was the reason to try and build one? Or was it an accident, like penicillin? Genuine question. Trying to get back to the original thinking.

LLMs in medicine are super exciting. I get it. Yay LLMs 🎉 🤌 Now, let's start doing the hard thing, and getting them through regulatory approval as SaMD. If you truly believe they are robust, safe and effective, prove it to the regulators. Until then, they are a just toy.

3 questions to ask about any clinical AI system: 1) Is the data immutable? - is the data displayed always correct? 2) Is the system deterministic? - given the same combination of inputs, the same answer is always reached 3) Is the output reproducible? - can someone else

LLMs make stuff up. Don't rely on them in court (or in a hospital for that matter). https://www.bailii.org/uk/cases/UKFTT/TC/2023/TC09010.html

When is the right time to start health economic modelling for AI medical devices? Early! https://www.hardianhealth.com/insights/early-economic-modelling-of-ai-as-medical-device-answering-the-why

Actual footage of radiology AI vendors announcing their Large Visual Models (LVMs)

Ever wondered how to get your AI solution into Africa? Tomorrow morning at #RSNA2023 I'll be talking through the patchy regulatory landscape and best practices on importing medical devices to LMICs. Moderated by @RogueRad...

Key takeaways from #RSNA2023 - Very few new start up entrants to the AI space. - AI vendors have unilaterally downsized their booths - indication of funding? - A few new point solutions (CDS or diagnostic) from established vendors but no game changers. - Talk is all about

Has anyone spotted a health economic model for an AI solution at #RSNA2023 yet?

Globalisation of medical device regulations is obviously something everyone should be on board with. However - it's not easy, as demonstrated by in-fighting, accusations of copy/pasting and fears of overt industry lobbying at the Global Harmonization Working Party (GWHP) The

The reality of regulation is that if we let 'big tech AI' go unregulated, then the onus of proving compliance will be on the end developers, application developers and users. Who would you rather be liable for demonstrating safety & compliance? You, or the big companies?

Great paper on regulating AI. The authors argue that AI should require a licensure model akin to GDPR frameworks of best practice. "Developers, vendors, and users of such models should bear the burden of proof to justify why their AI system is not illegitimate"

Chief Exec of NICE @SamRobertsNHS on generated AI evidence being used for HTA submissions "We're not okay with those. We currently do not accept Al-generated evidence. I mean, that's a blanket statement, but we know that's where the worlds' going."

My @thetimes interview with @dannyfortson about AI medical device regulation in health, and the collapse of Babylon. https://podcasts.apple.com/gb/podcast/danny-in-the-valley/id1233991021?i=1000633552154