I went to the doctor complaining of constipation. He sent me for a CT scan which showed that I was literally full of shit. The prescription was for a large dose of MiraLax. Now I wonder if the risk of the CT scan was really justified given that plenty of people already tell me that for free and without radiation.
A. Newer CT scan machines use lower radiation doses.
B. If you're getting only one scan a year you're fine and within yearly limits of radiation dosage considered acceptable.
Remember that you'll get comparable levels of radiation even if you commute through the grand central station every day.
This paper is for lack of a better word, crap. It's becoming sensational for the conclusion it makes and I'm afraid it's now going to create more harm because of that.
> Remember that you'll get comparable levels of radiation even if you commute through the grand central station every day.
Gemini says this:
> A single typical CT scan delivers a dose that is roughly 1,000 to over 5,000 times higher than the dose you'd get from spending a few hours in Grand Central Terminal.
Was it hallucinating here, or are the commenters hallucinating? What OP is saying is just not true. A CT scan and normal daily commute in Grand Central station are NOT comparable in terms of radiation received. Somehow this is controversial because an AI said it?
The machine appears to have hallucinated the incomparable comparison, instead of a human.
(And I'm not picking on the machine at all here. I use it all the time. At first, I used to treat it like an idiot intern that shouldn't have been hired at all: Creative and full of spirit, but untrustworthy and all ideas need to be filtered. But lately, it's more like an decent apprentice who has a hangover and isn't thinking straight today. The machine has been getting better as time presses on, but it still goes rather aloof from time to time.)
What do you mean all LLM output is hallucination? Would you say the same about AlphaGo? That system was also trained to predict human data initially yet it's competent to the point of beating most humans in Go.
> Weird you don't have this requirement for the OP spewing his urban myths above.
It isn't my purpose try to convince you that your apparent presumption that the output of a human and a machine are somehow equivalent and should be treated equally is wrong.
Why is an LLM more prone to hallucination than AlphaGo?
> It isn't my purpose try to convince you that your apparent presumption that the output of a human and a machine are somehow equivalent and should be treated equally is wrong.
You should judge arguments by their merits, not by who is saying them.
This data is from 2006. Over 20 years, there has been substantial progress in CT radiation reduction through model-based iterative reconstruction and now ML-assisted reconstruction, aside from iterative advances in detector sensitivity and now photon-counting CT.
In clinical practice, those doses are about 2-3x what I see on the machine dose reports every day at my place of work.
In thin patients who can hold still, I've done full-cycle cardiac CT and achieved a < 1 mSv dose. We are always trying to get the dose down while still being diagnostic.
Someone I know got partial constipation, lack of apetite and stomach reflux. After some month of triying to fix the reflux, they discovered it was intestine cancer when the cancer was big enough to block almost all the flow.
IIUC from a sibling reply, you already used a few laxative, so perhaps a CT scan was the next step.
I had a CT scan ordered for abdominal discomfort which had been making sleep a nightmare for years.
I started taking ag1 and Metamucil and the stomach discomfort has completely gone away over a span of weeks/months.
The resulting dramatic improvement in sleep cascades to just about everything in my life and different recurring health problems keep disappearing now.
It’s crazy how many problems can be caused by apparently just not getting enough fiber.
Never ended up getting the CT scan, which probably would’ve been expensive and involve some small degree of risk.
But did those words really get through to you like an energetic photon? I think, it's quite ionic how frequently the last strand has to break before we realize, we should have been radicalized by people, not high voltage discharge events.
I did try a boatload of OTC products, and eventually went to a generalist and then a specialist when it became severe and chronic. The OTC products recommend that on their labels. I now manage it with careful diet and a daily dose of polyethylene glycol.
If its any consolation, having diagnosed a malignant papilary or follicular thyroid tumor (surprizingly) is correlated with equal or slightly increased lifespan compared to the population that did not have this tumor—maybe it is because of having more frequent or better access to healthcare, which is harder to control in such epidemiological studies. It is one of the few positive known relationships with cancer.
Yup, this is something that shows up¹ for melanoma, thyroid, and prostate cancers: those who are diagnosed when the cancers are in situ (haven't left the tissue of origin) have better 5-year survival chances compared to people of the same age, race, and sex. Likely because, if people are diligent enough about their healthcare to report early warning signs and get the recommended screenings, those people also have much healthier lifestyles. If there were a way to control for lifestyle, I'd imagine the "benefit" would disappear or become a small but clear negative.
1: You won't find relative net survival above 100% in the CDC's statistics. That's because they calculate survival rates using daily differences in death hazards derived from life tables of people with cancer and those without. Add up the differences across all days, do some exponential math, and voila: relative net survival rates. But, if the relative risk for a day is negative (i.e. those without cancer have a higher risk of death), then they set the relative risk to 0 instead for that day. Which is ridiculous, IMO. It's forcing a distribution of actual events to match an idealized model.
Only after the procedure my ENT told me that if she were to pick a cancer to get it would be this one, and that the experts are trying to rename it from cancer to neoplasm because of its relative benignity. But I hadn't heard that it could actually be a positive. That underlines the mixed blessing of screening tests.
It also underlines the second-order risks of unneeded diagnostic screenings, even those that don’t include ionising radiation and its ilk, as the surgery you may otherwise have not had is itself not free from risk. Of course the case generalises out from there as well to not only this specific cancer.
I knew the math wasn't limited to 90 degree slices. I've had a few CT scans, and twice when I got my copy the DICOMs were a bunch of 90 degree TIFF slices. It wouldn't surprise me if the radiologist had the raw beam data and could slice it however they chose.
the Tiff slices is just post process storage. A good DICOM reader can view those tiffs in any direction.
After all, if you have the tiffs, you already know the grayscale for every pixel in the XYZ space. You just need to grab the right pixels from each slice
Interesting stuff! Could you suggest a DICOM reader that does so?
I had assumed slicing the raw beam data would give you much better output resolution whenever there were differences between transaxial and z resolution, since at some angle and offset you end up sampling across the largest gaps in the XYZ pixels. But maybe that's not a real issue? Is this how it's done in practice?
