I’m a physician in the US. In my experience main cost drivers are administration (documentation is part of it but the way billing works is you basically do the job, document what you did, then submit to the payer. This process can take 6+ months and they often just don’t pay for parts of it).
And then ICU/end of life. Way too many Americans don’t ever talk about what will happen when they die and what they do or don’t want at the end of life. Then they become incapacitated, family pursues all advanced therapies because they don’t know what to do and they spend a long time in the ICU getting things done for an unclear goal
Thanks! I half-jokingly want to come up with a mental health/memory + physical health test for myself where if I ever fail it my wish is to not be a burden, don't extend my life.
Radiologist. I don’t read MR shoulder exams in my day to day practice, but from the few pictures shown , I can’t conclusively disagree with the original report.
These models are generally terrible at reading medical images. The amount of public training data on the internet compared to the number of scans a radiologist reads in training is minuscule. There’s obviously a ton of medical images in general but very few, and even fewer along with a report are available on the internet publicly for download.
There are vision language models coming out of research labs that are excellent in describing and localizing findings. Still at the level of a 1st or 2nd year radiology resident, but as we all say - this is the worst the models will ever be.
Absolutely. It's very unfortunate that this post used the worst example possible of using LLMs for medical purposes.
General-purpose LLMs are _fantastic_ at medical diagnosis that do not involve imaging. I am completely convinced that given enough information and time, frontier models already outperform >90% of doctors on initial diagnosis of internal issues and suggesting medical tests to further reject or confirm the most likely theories. To the point where I'm eagerly waiting for the first hospital in the world that's willing to be open and honest about using them for that first step, and then proceeding from there. I'll be on a flight there as soon as one arrives.
At the same time, they're worse than useless at anything involving medical imaging. Asking them to interpret them is worse than trying to interpret them yourself as a layman. And you surely wouldn't interpret them yourself.
It's multiple things. It never shows the subscapularis in the way that people actually look the tendon. It hyper fixates on the axial when I find the sagittal much more useful for subscapularis.
Figure 7. There's an arrow pointing "to the acromial undersurface". The arrow is not pointed to that location.
Figure 5. "thin bursal fluid". This is within physiologic variation, but is calling bursitis.
It keeps bringing up irrelevant normal things like the shape of the coracromipal arch, I assume because lots of websites have information about that as a patient focused possible cause for rotator cuff impingement.
I am reminded of the recent Stanford MIRAGE study which found that LLMs will happily hallucinate answers about medical images if the medical images are omitted.
I don't understand why this is still confusing to people. The second "L" in LLM is language; these things are AWESOME at producing things that SOUND like language, including code. They have so much training data that it is almost always grammatically correct, and often makes sense. Extending this, it has obviously been trained on data containing phrases like "acromial undersurface" and "thin bursal fluid", and "coracromipal arch", in the context of shoulder injury and related imaging. BUT IT DOES NOT KNOW HOW TO DIAGNOSE ANYTHING. So, it SOUNDS like a radiologist or specialist, and might be in the ballpark of correct-ish-ness, but ultimately is a fancy Markov model.
> Can you share the reasons that you believe this?
Firstly, please keep in mind I'm talking about the entire doctor population of the world here. Not sure which particularly bubble of this earth you have experience with, but note how half the word's population lives in India/China/Indonesia/Pakistan/Nigeria/Brazil/Bangladesh/Russia. Now I do believe that it holds the same for e.g. Europe and non-China East-Asia, but still.
How many patients has the world-wide average doctor seen? How long have they been a doctor?
How many have they seen with the particular condition the patient has?
How much time do they spend listening to and reasoning about a patient? The median in the world is likely under 3 minutes.
How many real-world incentives do human doctors have to deal with?
Given infinite time and resources, and zero external incentives, maybe the median human doctor would outperform the LLM at this task. But this is completely detached from the real world.
> What is special about medical imaging that makes AI/LLMs specifically bad?
LLMs: Besides lack of training data as mentioned elsewhere, they're simply not trained for high-fidelity image processing in general. It's not limited to medical imaging. It's a bit like the "How many Rs in strawberry" thing, but worse.
As for "AI" in general, medical image analysis is a very active field. These tend to be purpose-built though, not general-purpose. It seems likely at some point they'll become mainstream, but there's still a way to go.
Yeah, medical computer vision is a (fascinating) field with a lot of ongoing research. SOTA models are highly specialized, and are only getting good enough to be used by actual doctors and patients. Using a general purpose LLM to do this is similar to giving a credit card to Openclaw and telling it to make you rich through the stock market & cryptos.
Somewhat. During residency I developed models for detecting liver tumors in MRI imaging. Like you said, highly specialized and a lot of manual work developing the dataset.
There are now open source open weight “foundation models” coming out of labs that are transformer or mamba based architectures. These will accelerate development.
I don't have insider information, but: if one of the AI companies really wants their models to become really good at this and publicly available datasets are scarce, they can probably just buy anonymized X-ray/MRI scans paired with the human doctor's diagnosis, and train on them. I don't know what the legal story is around this, but AI companies have near infinite money, so I'm sure they can buy their way around regulations (eg. by buying them from a less regulated country).
My understanding is that medical images are part of a patients record (so they must be available for the patient or other docs at the request of the patient) but whoever has collected the images does have some form of ownership. I’m not a lawyer and I have a cursory understanding of this. I believe it would be possible for an AI company to lease or get access to the data through a transaction but I suspect that it hasn’t happened (or happened publicly) due to fear of backlash.
For example I know that some companies like Tempus had access to imaging that corresponded to tumors which had been biopsies for sequencing and they were developing models in house.
Anecdotally, I've had Claude (Sonnet and Opus latest) consistently misread numbers from screenshots of my macro tracking app. Makes me skeptical of claims about its usefulness for anything requiring accurate image interpretation, let alone MRI analysis.
No trolling here: Do you feel threatened by the advance of AI/LLMs with respect to your field? I would. I am a computer programmer, and it absolutely feels threatening.
I mostly do interventional radiology, which is more similar to a surgical specialty than a diagnostic radiology practice. We do a lot of procedures , see patients in our clinic, have a service that rounds on and follows up on patients etc. I don’t think AI will affect my job much in the next 10-20 years. Advancements in AI and robotics could potentially offload some of our work, but most of what’s out there is underwhelming, but we all know that can change fast.
Even for diagnostic, I’m not totally convinced AI is going to cause a lot of issues. From a medical-legal perspective I doubt AI companies are willing to take on the risk of misdiagnosis . When that happens, human rads will start to get phased out.
In the interim, I think the next 10 years could be a golden period for diagnostic rads. Rads will still be the ones doing the work and signing reports, but people who learn how to use the right combination of tools will become very productive and can make a great living. Eventually payers will re-align but early adopters who figure it out will have a leg up.
As a programmer, I don’t feel threatened by the technology itself, but I do feel threatened by the second-degree effects such as what the technology does to our field, especially in the wrong hands.
I'm just guessing here but similar to a CAT scan, having actuators/probes at all angles could mean you can get an image around such obstacles. skull is probably an exception and it's the reason why we don't see any head scans in any one of the videos.
If you could obtain volumetric/3D ultrasound data that was not operator dependent, that would be great.
US is a good diagnostic tool, but it can be challenging to read because obtaining good images is very operator dependent. You need to have a good sonographer that can get the right views, knows how to adjust the imaging parameters to produce high quality images. It's not like CT or MR where the tech just sets a few basic scanning parameters and let the machine do its job.
However, see my other comment, the example images they provide on the page do not look great, very limited organ detail.
I'm scratching my head about why they would venture into an entirely different field like this, one with tremendous regulatory hurdles, if they know (and surely they must know) that radiologists are going to pan the results.
It's like if LeBron announced he was switching to bowling and was going to revolutionize the sport, then rolled a gutter ball.
Not sure. Image reconstruction/generation is a computationally intensive process, and in recent years DL based methods for improve image reconstruction have advanced fields like musculoskeletal MRI imaging. The physics behind this idea are interesting, but will have to wait to see if they produce images with high anatomic detail.
Instead of the value of evaluating a single scan, what about determinations made from evaluating regular deltas between images?
As a layperson, I'm mostly familiar with the concept of "get scanned, and a professional evaluates it"... are there scenarios where the approach of "imaging every few weeks, to make decisions based on trends" is currently done?
(From reading other comment threads here, I suspect the general answer is: other body-scanning startups have proposed the same thing, and it hasn't made sense)
As an aside, I could probably benefit from allergy shots, but the idea of having a regularly scheduled errand to do during the workweek is pretty unappealing, so I never seriously consider it.
Besides the high probably that those images are fake, and probably this entire device is fake... if it were real then it would mean what they're showing in those images is not even close to an approximation of what the actual data could show you if they put more effort into volume rendering of 3D data (not unlike Voreen).
The resolution of typical DICOM images is much less than what they're saying they are actually capturing, so the reconstructed images they're showing are just terrible for no good reason.
But I suspect there is a bigger fundamental physics issue with this entire thing... I'm not convinced they can penetrate fully inside and all the way around a human with only non-ionizing energy, especially from that far away.
can you say more? dont look great compared to current radiology, sure, but you see no potential in ultrasound diagnosis whatsoever? would it improving 10% change your mind? 10x? what's a good way to think about what "looks good" looks like?
That's basically the only thing I'm interested in reading about this. Based on my complete lack of radiology knowledge, I'd say the images look... a bit blurry or something? So, what would be an example of something this would not allow a radiologist/doctor to see?
Without those kinds of details, radiologists just expose themselves to: oh so you're telling me this doesn't work as well as the machines you paid ~millions of dollars for and are currently charging your clients a lot to use? Mmm I wonder why.
More data sounds better, but especially in a medical context, you have to be careful, because false positives have consequences. The PSA test is no longer broadly recommended for prostate cancer screening [1]. What harm could it do, you know more about your body, even if it's a noisy predictor? Most prostate cancer is slow growing, and something that men "die with" rather than "die of", so treatment can make for worse outcomes, without clear benefit.
It's not clear that we have the health infrastructure in place to know what to do with frequent, low resolution, whole body scans of the human body. How often do anomalies show up and then go away? How often are anomalies purely a scanning/data processing artifact? Who reads the scans and makes recommendations about follow-ups, if any? I think this is the kind of thing that sounds exciting and with low direct risk, but with all kinds of questions that are not only unanswered, but apparently unconsidered.
> It's not clear that we have the health infrastructure in place to know what to do with frequent, low resolution, whole body scans of the human body.
This is exactly my thinking. There are decades of longitudinal studies behind the recommendations physicians make based on given levels of e.g. cholesterol in a standard blood test. And critically, those depend on standard protocols around administering and testing samples.
This would be brand new and would not have any of that infrastructure. Which all tech starts at, good. But I would expect Midjourney to need to dig in for a few decades to get and analyze clinical results and outcomes.
For body scans, I think about how few people would know if they have e.g. three kidneys (or other distortion), and how that impacts/doesn't impact their health.
Most people do not undergo autopsy after death, so it's possible there are correlates between good/bad health outcomes that frequent scanning would eventually reveal. But it would take significant time for this to be apparent.
Yes. I spent a bunch of money on many of the optional extra imagining scans on my last health check up only to realize this afterwards. Humans have survived this far without this data. It would be better to spend resources on preventative things or lifestyle things known to promote health, than to obsess over seeing whats going on inside.
Other than the shapes of the tissues in the images, there is no anatomic detail. Wouldn't be useful for diagnostics. It's substantially worse than conventional ultrasound.
Would it be suitable for basic body composition (as they claim in TFA)? DEXA is a big business and companies push a subscription model where they encourage you to get monthly scans. The results are really fun to look at and the dose is admittedly very low, but you're still getting rastered by an x-ray. It would also explain the spa angle and hence why they're doing that before going for regulation.
> We’re starting by just giving you detailed body composition maps — and we’ll be submitting regular test results to the FDA for increased capabilities.
As far as I understand ultrasound there's no reason you couldn't do this, it's just infeasible to do a full body scan with a hand probe and you get covered in goop.
I only did the $20/month subscription since 9/2025
It was great for about 5 months, amazing in fact. I under utilized it.
For the past month, it’s basically unusable, both Claude code and just Claude chat. 1-2 prompts and I’m out. Last week I prob sent a total of 15 messages to Claude and was out of daily and weekly usage each day.
I get that the $20/month subscription isn’t a money maker for them, and they probably lose money. But the experience of using Claude has been ruined
There is a lot in this comment I agree with, however I think may universities have backed themselves into a corner with the degree of tuition inflation that has taken place over the last 20+ years.
I graduated from a SUNY school in 2012. At the time, you could still actually go to school and work part time and get through it. Not saying it was easy by any stretch but it was possible. Tuition + living expenses were about $17/year on campus , less expensive housing was available off campus.
Now, even state schools have tuition which is only affordable through family wealth or loans. Going to university is no longer a low stakes choice - if you flunk you’re stuck with that debt forever. Not to say students aren’t responsible for understanding that when signing up, but the stakes are just a lot higher than what it used to be.
Interesting but how would this prevent against “off-chain” collusion? A fraudulent seller captures brokers on the buy and sell side? Seller backs out of the deal unless they know who the brokers are?
I think this kind of behavior in principle would be detectable but in principle with enough concentration in the market, a fraudulent seller could in practice get brokers and jurors to collude with them.
Once a trade has been initiated, sellers won't be able to back out. The seller has no control over the pseudo-random broker selection for their trade, hence can't choose a preferred broker either.
A few points based on comments I’m seeing about the article.
This method of ultrasound treatment is called histotripsy. The underlying mechanism it uses to treat tumors is by focused ultrasound beams that mechanically disrupt cell membranes . It basically turns the lesion into soup. It does not treat the lesion by heating, although there are other techniques that do use ultrasound to ablate tissue with thermal energy.
Where I have seen it used and discussed is in the liver, whether that be metastatic disease to the liver or primary liver tumors.
One challenge is that in the liver you can’t use it for lesions that are near the capsule of the liver. It can also be difficult to keep the ultrasound beam focused on the lesion with respiration, especially if the tumor is small.
It’s an interesting technique and I think more people will use it over time. Whether it will be better than other established techniques like microwave ablation or radioembolization (for liver tumors) remains to be seen. I’m an interventional radiologist.
And then ICU/end of life. Way too many Americans don’t ever talk about what will happen when they die and what they do or don’t want at the end of life. Then they become incapacitated, family pursues all advanced therapies because they don’t know what to do and they spend a long time in the ICU getting things done for an unclear goal