Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Midjourney Medical

midjourney.com

Post reply on HN
Print view View original

481–490 of 927 posts

Re: Midjourney Medical

#481

Post by nickpp »

camillomiller wrote: post #464
wkoszek wrote: post #382

Earlier quoted context omitted.

All doctors say this, and that sort of drove me away from healthtech. As if there were absolutely no way to take a step in a direction of fixing it. The faster and earlier we start to scan everyone regularly, as long as scanning methods aren't invasive, the more certainty we'll have what to warn people about and what not to tell them. Perhaps with the regular screening (imaging quarterly, if the scan is fast) you cou…

This Silicon Valley mentality applied to a mechanistic view of the body is a fucking disgrace. This will fail, and luckily we won’t have to endure more of Silicon Valley’s dunning krueger on steroids about medical solutions. The Silicon Valley has NO CLUE of the complexity of clinical science, yet they hold this populistic view that everything can be foxed with tech and nothing stops the hybris. We all can see where…

Aren't most of the current/latest advancements in health care coming from tech and software?

Re: Midjourney Medical

#482

Post by dmurray »

luesterklemme wrote: post #466
dmurray wrote: post #454

Earlier quoted context omitted.

MRI machines cost in the six figures [0], last 10+ years and could reasonably do thousands of full-body scans a year. That's basically free by healthcare standards. Rent for the room to put it in would cost more in most cities. MRI operators are specially trained technicians, because these are complicated machines. But like, semi trucks and photocopiers are fantastically complicated machines, and we seem to be able t…

What are you reasonably expecting to find in a full body MRI? Besides the notion that a "full body MRI" is not a procedure that is routinely done anyway and lasts upwards of an hour. It's not the scanner that is the limiting economic factor.

Right. I'm replying to the commentator who questioned how we could possibly purchase and staff enough MRI machines to give people regular full body scans.

I'm saying there's no question that would be economically viable. The reason we don't and shouldn't do it is that it wouldn't be medically valuable, even compared to other cheap interventions.

Re: Midjourney Medical

#483

Post by andreareina »

KingMob wrote: post #413
wkoszek wrote: post #382

Earlier quoted context omitted.

All doctors say this, and that sort of drove me away from healthtech. As if there were absolutely no way to take a step in a direction of fixing it. The faster and earlier we start to scan everyone regularly, as long as scanning methods aren't invasive, the more certainty we'll have what to warn people about and what not to tell them. Perhaps with the regular screening (imaging quarterly, if the scan is fast) you cou…

It's more statisticians saying this, and not doctors per se. You run into issues of signal detection theory, false positives, and the lay confusion that Bayesian P(A|B) !== P(B|A). You're right that we could take steps to fix it, but unfortunately, those steps involve mass education that every human body has anomalies, and many of those should just be ignored. We'd get a wave of anxiety, lawsuits, and unnecessary int…

It's also doctors. Medlife Crisis on YouTube, Barbell Medicine, others. BBM have an article on priorities for overall health and they link to a tool maintained by one of the professional bodies on what routine screens to have done and it's pretty conservative. Even my doctor on seeing an "abnormal" lab result said it was likely spurious given my lack of complaints and all the rest of the results. That said they still recommended a follow-up because they kind of have to given professional ethics. BBM (again) made a similar point: resistance training is known to cause liver-associated enzymes (AST, ALT, etc) to rise, that doesn't mean you can ignore a high value.

The steps to fixing it is to not take the test that takes you from a prior of 1/100000 to a posterior of 1/1000, because you're going to ignore it anyway. And you can't depend on multiple testing because those test results can be correlated.

ETA: I can be convinced that we can collectively get to a place where broader screening would be indicated. But I think it's going to require both of the tests getting better and being better about what we do with (and feel about) the results.

Re: Midjourney Medical

#484

Post by newsclues »

convnet wrote: post #287
tgsovlerkhgsel wrote: post #233

I think a lot of medical diagnosis could be solved with mass data collection if it was cheap enough. Right now, blood draws are somewhat routinely done because they provide a lot of human-interpretable indicators from a small number of values, and there is some evidence that e.g. "dogs can smell cancer" etc. (i.e. some diseases cause detectable odors). With a big enough data set of [all kinds of bio values, including…

It's a controversial and complicated idea. The downside, and the reason why most doctors do not recommend full body scans, is that every human body is a bit weird and there will almost always be something "wrong" that will be visible in a full body scan. This can lead to unnecessary testing, anxiety, and even unnecessary procedures. Many of these oddities flagged by the scan would never have caused any actual issues…

All the reasons you’ve listed are excuses why my government healthcare stopped having annual checkups. But to me it’s just worse quality care

Re: Midjourney Medical

#485

Post by looofooo0 »

haldujai wrote: post #298
adastra22 wrote: post #262

Earlier quoted context omitted.

We have no evidence in favor of the linear no threshold model. That is not the same as saying that we have evidence against it.

There is some evidence for hormesis - but yes no model is proven right now. LNT is the most conservative model and part of why it sticks around. A good primer: https://pmc.ncbi.nlm.nih.gov/articles/PMC2477686/

LNT does also damage, as people refusing necessary CT scans or countries switching of nuclear power because of fear.

Re: Midjourney Medical

#486

Post by friendzis »

wkoszek wrote: post #390
friendzis wrote: post #336

Earlier quoted context omitted.

There's a ton of variation within medical testing and tons of different conditions affect test results in similar ways. VERY FEW tests (test classes maybe: biopsy, microbiology, skeletal Rö) can yield diagnoses in the first place. Most testing is used to support (not confirm!) and reject possible interpretations. This non-invasive everything-scanner sounds more like science fiction.

Other tests should be solved too (fecal/urine/blood). Perhaps we need more R&D in here to accelerate progress. We already have patients trying to track their own health over longer time which is great. We then just have to make AI good enough to spot warning signs (without patients asking). Or parhaps we need to make those tests easy and cheap and regular.

> Perhaps we need more R&D in here to accelerate progress.

In general yes, just that "more" is monstrously massive to the point of it being closer to science fiction than reality, IMO.

To reiterate, various assays fluctuate rather wildly over the course of various body cycles. The reason(-s) your doctor should remind to get a blood drawn in the morning after a period of fasting is that the sample is taken at a somewhat steady state and the result is comparable to reference values without too much of a margin.

Anyone with a requirement to manage blood glucose levels will tell you that CGMs are vastly superior to finger pricks first and foremost due to the sample rate available and comfort reasons secondarily. With a finger prick test the patient is only somewhat aware where in the curve they are, which makes the test only a rough estimate due to this temporal error margin. A lot of people do not zero in their readings with finger pricks as they are mostly interested in the deltas.

Suppose you manage to make urine sampling relatively accurate and super cheap (e.g. tens of eurodollars per analyzer or cents per test strip) so you can have poorly-supervised, long-term studies with huge cohorts. However, unless you somehow control for sample collection conditions, all this baseline variability suddenly infects your whole dataset and effectively raises noise floor. It's not unreasonable to expect that whatever was found to be a useful signal under controlled conditions to fall below noise floor under uncontrolled conditions.

That's basically THE problem with the hypothetical test-it-all machine. Again, maybe in some cases that could be extremely useful, but in a lot of cases that would be counter productive. However, what CGMs hint us at is that various kinds of Continuous X Monitors could provide insights into body reactions to things, which is, currently, effectively not a signal in general medicine. Once the test-it-all machine is reframed as an array of continuous monitors and the useful signal is reframed from long-term drifts to short-term deltas it may unlock some additional diagnostic pathways.

Re: Midjourney Medical

#487

Post by adastra22 »

jibal wrote: post #342
adastra22 wrote: post #259

Earlier quoted context omitted.

Only if you let them. The false positive thing is a nonissue that only arises from assuming you would respond to information a certain way.

That makes no sense at all, unless you're saying that people should respond to all such information by ignoring it.

You could repeat the test, perhaps on a more frequent interval to keep an eye on it. You could follow up with a more specific test, or do confirming blood work. In the meantime you can adjust your diet as a precaution, or get motivated to get in shape just in case.

There's plenty of room between "go under the knife" and "ignore altogether."

Re: Midjourney Medical

#488

Post by EEBio »

tgsovlerkhgsel wrote: post #233

I think a lot of medical diagnosis could be solved with mass data collection if it was cheap enough. Right now, blood draws are somewhat routinely done because they provide a lot of human-interpretable indicators from a small number of values, and there is some evidence that e.g. "dogs can smell cancer" etc. (i.e. some diseases cause detectable odors). With a big enough data set of [all kinds of bio values, including…

There is this infamous DANCAVAS study [1] in which they ran cardiovascular screening on older population and found absolutely no benefit in doing so. Note that while the linked study claims there was a benefit to screening (reduced mortality in men aged 65-69), the apparent effect was caused by improper statistics (primary endpoint did not show benefit, only subgroup analysis, which however lacked statistical power to be reliable). And indeed, their follow up study showed that there was no benefit in the subgroup. [2]

Many a dollar is wasted every year on trying to prove population-wide screening prevents mortality or increases patients’ quality of life and every time we don’t cheat with statistics we get the same answer - population-wide screening isn’t effective.

1: https://www.nejm.org/doi/full/10.1056/NEJMoa2208681 2: https://www.escardio.org/news/press/press-releases/No-signif...

Re: Midjourney Medical

#489

Post by looofooo0 »

unholiness wrote: post #300
arcticbull wrote: post #271

Earlier quoted context omitted.

Sure but we don’t prove negatives for a reason - it’s impossible. We assume the null hypothesis.

LNT is the null hypothesis. No one disagrees a linear model fits the data very well in high doses. If you want to argue that model doesn't work in low doses, you need a model with more parameters and sufficient data to fit it. The issue is that, at these low doses we want to differentiate, we're also looking at effect sizes that are hard to separate from noise, and sampling biases that are hard to erase. There's stil…

Well problem is that humans are so noisy through lifestyle, enviroment and genes that any proof for either is really hard.

Re: Midjourney Medical

#490

Post by rzwitserloot »

convnet wrote: post #287
tgsovlerkhgsel wrote: post #233

I think a lot of medical diagnosis could be solved with mass data collection if it was cheap enough. Right now, blood draws are somewhat routinely done because they provide a lot of human-interpretable indicators from a small number of values, and there is some evidence that e.g. "dogs can smell cancer" etc. (i.e. some diseases cause detectable odors). With a big enough data set of [all kinds of bio values, including…

It's a controversial and complicated idea. The downside, and the reason why most doctors do not recommend full body scans, is that every human body is a bit weird and there will almost always be something "wrong" that will be visible in a full body scan. This can lead to unnecessary testing, anxiety, and even unnecessary procedures. Many of these oddities flagged by the scan would never have caused any actual issues…

One obvious alternative plan, presupposing that Full Body Scan is dirt cheap, is the following protocol:

- At 25 years old or whatever you get a FBS. Pretty much no matter what, this FBS will not be used to do more checks, procedures, and so on.

- ... and now we give you another FBS every so-many years, and only those things that are different from the previous scan are investigated.

There's still an issue with needless procedures, but the amount of 'weirdness that are not going to cause an actual issue had the patient never been aware' is significantly reduced by looking only at changes. i.e. most 'weirdness' shows up early and is fairly stable.

The difficulty is the moral issue. You cannot show that first scan to the patient. Even if every soul agrees beforehand that the rule is that nothing on that first scan, no matter how scary it looks, is further investigated... any medical issues raised by patients are used as a major information input for diagnosing issues. If I show a patient a scan that has this tumor looking thing on the left lung, then no doubt a few months later they'll be back complaining about shortness of breath and a pain on the left side of the torso. The mind is a powerful thing. At that point you can do a scan and see... the same nasty tumor looking thing we saw on that first FSB, and we're right back to the issue of these scans doing more harm than good.

Is it morally acceptable to hide that first scan from the patient?

Post reply on HN