“Amateur” programmer fought cancer with 50 Nvidia Geforce 1080Ti
howardchen.substack.com
howardchen.substack.com
Here is the link to the original article: https://www.toutiao.com/article/7094940100450107935/
Seems pretty effective. :)
What a beacon of light and inspiration you are. Thanks for your work.
That said, I welcome you to publish your work so it can become even better after a formalized peer-review process.
I'd love to chat.
1. do you have a patreon or something where people can support you?
2. do you know how to obtain the nucleotide sequences of c-MYC?
3. have you seen machine learning applications that can predict protein function? specifically, the goal is to predict if protein X could participate in cleaving of DNA similar to CRISPR Cas9? false positives are okay, provided the negatives are accurate (i.e., guarantees protein cannot cleave DNA like Cas9).
Check:
https://www.ensembl.org/Homo_sapiens/Transcript/Summary?db=c...
This is the main transcript, the remaining 9 are in the table
for instance, how to obtain the nucleotide sequences for TP53, or for the NHN/RuvC domains of CRISPR Cas9?
thanks again for your help.
same site (ENSEMBL): https://www.ensembl.org/Homo_sapiens/Gene/Summary?db=core;g=...
also you can check human genes at GeneCards: https://www.genecards.org/
re other questions: The crucial thing is: what are your goals? Bioinformatics and cancer biology in particular are rather vast fields. Questions and answers make sense if you can place what you just got in some framework. And getting there is non-trivial, see i.e. +800pp book: https://www.amazon.com/Lewins-GENES-XII-Jocelyn-Krebs/dp/128...
is it possible to find nucleotide sequences from amplified c-MYC from cancerous tissue, either burkitt's lymphoma or TNBC?
ensembl is valuable, but it would be more efficient to hire someone to decode the different options. what type of researchers understand the different sequence options on an ensembl page: cancer biologist, geneticist, bioinformatics scientist?
thanks again for your help!
the goal is conducting some first-principles analysis of cancer, nothing fancy and very likely a waste of time.
Whole brain radiotherapy works by killing everything a little bit in the hopes that the tumors die first (e.g. like chemo). There are good reasonswhy this tends to mostly-sort-of be true, but getting the balance right is hard and too much dose will definitely cause other problems.
SRS is a way of targeting radiation directly to locations to kill cells, with less effective dose (hence damage) to other parts of the brain.
It's all pretty harsh stuff, and you can die from the necrotic tissue caused by it, also.
Often with this kind of disease you know you aren't going to cure someone, but you can get rid of symptoms and make people more comfortable (palliative care).
Not least, Nvidia's documentation is not the best resource to learn from. This seems like quite a lot of work to understand ML and write custom CUDA code to get this to work. Do you have any insight about how you taught yourself these things and what tools you use?
Usually you would - nowadays at least - use someone else's optimized kernel + ML library but if you wanted to roll your own that's doable.
It's a nice thought exercise, but of course not realistic. FAANGs wouldn't be able to attract or retain the talent if they were purely in scientific medical research (where do you think those fancy offices and fat paychecks comes from?).
Having said that, it doesn't have to be either/or. I strongly believe that companies can be successful in balancing between thriving commercial businesses and using its profits to advance research that can benefit the society at large. Case in point: Google's DeepMind has for years published groundbreaking work in various areas including breast cancer research [1], protein folding [2], energy savings and climate change [3], and more. Also Calico [4], or its investments in health startups via GV, and of course all the well known services to users.
What I'd like to see is more companies following a path where a portion of its profits is used to fund research in areas that can cause societal change for the better (which, in its turn, may eventually became new sources of revenue down the road - e.g. protein folding, self-driving cars).
(disclaimer: Googler here, but no relationship with any of the projects mentioned)
[1] https://www.deepmind.com/publications/international-evaluati...
[2] https://www.technologyreview.com/2021/07/22/1029973/deepmind...
[3] https://www.deepmind.com/blog/real-world-challenges-for-agi
Calico (California Life Corporation)
(I’m hoping to work there after my PhD)
A decade ago I did a voluntary work on a simple app for oral cancer detection(questionnaire, data-entry) with an oncologist, Who used it to for survey in the tribal regions of India and he used to say how lack of early detection is the number one reason for so many deaths(many die without knowing that they had cancer).
He's now settled in Germany, But I would still pass this story to him & perhaps his colleagues in India could make use of it.
Thank you.
I have a question for you. What is the tech stack that you use?
And if it is not too much: What resources did you use to learn Deep Learning?
I apologize if this has been asked and answered before, but do you speak Mandarin, or was the interview in English?
Asking out of curiosity if it’s the former, and if so, how difficult was it to learn whilst also working on this and other things? And are there any resources or tips you might share that you found helpful?
That's why not one startup has hacked healthcare in America, not one. No breakaway successes making pharma cheaper. Like those incubators in Bangladesh, for premature babies not startups that is, those did OK. Some pill startups yes, but again that's an expensivification of medicine. If you can make medicine more expensive, they welcome you in!
Jim Clark tried this, he was on a roll after Silicon Graphics and Netscape. Huge roll about as strong as Elon Musk as a serial entrepreneur. Then he targeted healthcare and couldn't do shit, just couldn't get anything to happen. He literally talked about getting "rid of all the assholes" by which he meant insurance and doctors and hospitals and middlemen and pharma and all the other "assholes" of that nature in his own words, but leave "only one asshole in the middle--us [paraphrased]." It's in a book. That book also talks about guys going on airplanes and chasing goats off cliffs, saying "Some people do this."
Well the real structure of medicine isn't designed around the human body, it's designed around cornering the market. Market dominance. So of course it has this immune system against cost reduction and efficiencies--efficiencies especially--and you do know it lobbies, don't you? And can bribe the FDA like the Sacklers did? Or lobby the FDA, and then bribe underneath so when people see favoritism they think it's the over-the-counter placebo causing a placebo effect without suspecting an additional more potent under-the-table dosage of money. In case the administration has built up a tolerance to the over-the-counter stuff.
This post makes a lot of points, but in general I think they boil down to the above statement : the belief that large, complex systems are just run by stupid and/or malicious people and that a sufficiently clever 'hack' will fix all the problems. I think that is an attitude that is common on HN, but wrong.
Most big problems are not technology problems, they are People problems with a capital P. Technology problems can be fixed with 'one simple hack they don't want you to know about!!!' People problems are complex and messy and cause and effect can be intermingled vertically and horizontally with other seemingly unrelated factors as well as temporally with things that don't even exist yet or used to exist but don't anymore!
The way we fix these messy, complex People problems is by respecting that they are real problems, that the people acting on those systems are (mostly) reasonable people just doing what reasonable people do, and slogging through solutions a day at a time with the oldest technology around - political power. These problems resolve if you can get enough people to agree they need solved.
If America has dysfunctional healthcare there is still the rest of the world. Which might be good for Americans eventually as the tech will come across one way or another.
Which raises the question - why does the vast majority of healthcare tech development come from the US? (I included "development" to get around the out-sourcing of testing to China.)
Similar to silicon valley’s tech scene, you get migration of people who want to work at a level where that sort of capital is required.
Modern treatments are hugely expensive to develop, and they also tend to be very specific (ie. only 1 in 5,000 people might get the exact right kind of brain tumour for your treatment to be an option).
With only about 1 million all-cause deaths per year in the UK, that means your treatment for a specific terminal brain tumour might only have 200 patients per year.
The 'we just saved 4 hours of clinician time by using fancy AI' just isn't worth it if it only saves 800 hours of clinician time per year, yet costs millions to develop.
The fix is to roll this out somewhere with more patients (eg. China) and where trials are cheaper (ie. China).
No don't be like coolwulf, be coolwulf. Be him, do what he did, make the algorithms from your brain protect the algorithms that make up your brain. You know they're symbiotic right?
Transistor and neuron fighting together against destruction. Symbiosis.
EDIT: coolwulf might as well be John von Neumann, the last line of defense against brain cancer and brain surgery, those neurons reaching beyond the skull to make a change in the World for some part of that World to then reach inside that skull and protect it. The chip as the final garrison of the biological human brain. Ghost in the Shell. Ghost retreating from the Shell and regrouping, no ruling out whether it can triumph when it returns to reconquer its damaged Shell.
Money moves too many times, like the pebbles under the shells. So that's a deflection so people don't see who's serving whom. Every time someone thinks they can remember what's under a shell before it's shuffled, they move it really fast, and like with feints. And of course this doesn't work for real, hands aren't fast enough, so you gotta layer it with like something to injure their eye with a flash while they're looking, say a laser to the pupil, or maybe...
You could make a little trapdoor for the pebble provided the guy will not ask which shell the pebble was actually in. So deflect all questions.
But the end outcome is servitude comes from all over the world to an American banker, an American doctor, and an American executive. There's a few others. But the long tail of the American middle class? That was only necessary because those bankers, doctors and executives would never themselves (or with their sons) man the nuclear ballistic submarines, dogfight jets, aircraft carriers, CIA undercover agents, and especially infantrymen--never never never infantrymen. I've only heard of a single lonely guy becoming an enlisted man in a high-bullets-flying environment, when he could have gone anywhere--from Philips Exeter in fact. Anecdotal.
Yeah undercover agents get wrapped up and tortured, but infantrymen are bullet sponges. Riflemen, holy shit. WW2 riflemen, no let's not talk about that. At least spies there's a whole song and dance to the whole thing, part two of 1984, and speaking as someone who's stood up to torture, there's not the immediacy of the infantry where the brain surgery is much messier. Well on second thought it's not that different from getting rocks thrown at you, like I have this year...
Listen. So this is what's up. There's these guys, they eat all this work and don't do anything. Doctors don't do anything. They only use 5% of what they learned, the rest being a masochistic filter the elders imposed on the youngers to keep them from bringing them down to human reality with competition. So morally a doctor should only study for 1 year, not 20 (if not less, because of retention). That's five percent! Don't argue with that part, argue with something else!
And then plus surgeons, spesh brain surgeons which is what every medical student "wants to be when he grows up" as elder doctors put it, so the thing there is younger surgeons have much better manual dexterity than older surgeons, you can't do it past your fifties generally. What does that mean? Learn it faster! But no, they have to say "I studied for 20 years to tell you my bullshit" like come on, that's masochism. You spent 1 year learning worthwhile things and 19 years in the medical school "slammer" learning to be a good bitch, that's the whole game in medicine, hazing. Because otherwise people would say hey just pump out more doctors then the doctors have to say "nonono, you can't because it's soooooo expensive to make a doctor and America loses money [pretends to lose money] on training doctors and you have to be the absolute best of the best of the best to learn how to multiply 3-digit numbers".
It's so so stupid, especially because of this: it's all multiple choice. All of it. It's all multiple choice. No essays or explanations, just filling in black circles. I have contempt for using that form of testing alone because I'm so good at it with no effort. Like not even doing the reading. Show up "naked", a good enough mathematician should be able to learn from the test itself, from NOTHING, NOWHERE, and NOBODY. Destroy the integrity of these measurements. Get the test to leak the answers. Like I could do it in Chinese [I don't read Kanji, I can only make out Oracle Bones from 2600 BC] and be measurably better than a monkey. I don't generally brag about it because I have no respect for it, like not having respect for torture. But if that's the underlying reason some doctor claims to be better than me? If such a doctor says he's better than me, then let us agree we are not equal.
Marty Makary wrote a book a couple years back, The Price We Pay on the wicked knot of a problem that is US health care. For those mildly interested, there was a Peter Attia Podcast inteview a while back that covers the gist of it: https://peterattiamd.com/martymakary/
For those that want to get some color from some of the biggest problems from a clinician/practioner's perpective, I found some of these podcast episodes to be pretty great/eye opening: https://zdoggmd.com/podcasts/
There, it may content errors, since it's a machine translated copy what had to be telegraphed by the operator. Thank him, he's a brilliant guy.
The above is with regard to a well-funded and regulated screening program that presumably has much better precision/recall than this website. I wonder what the cut off age is for this website before the diagnoses cause more harm than good? 60? 70?
This is getting lots of upvotes because it's confirmation bias for the large segment of HN readers who believe that problems would easily be solved by a small number of brilliant technologists, if only it weren't for governments and big organisations with all their rules and regulations.
Moreover, I personally don’t have confidence in their ability to make those kinds of decisions, and I believe the abysmal performance of the NHS supports my view.
I certainly agree that central authorities can be better. But that's kind of a truism.
What alternative options do you have in mind? Admittedly I'm short on alternative ideas.
But if you insist upon perfect equality of health care access, I guess it can't be. Some central authority has to decide if anyone is allowed to treat you, regardless of what you've saved for the eventuality.
Perhaps that’s part of my developing thought on the matter: the public system paid by tax dollars should be equitable to all. But by all means, take your credit card to Pepsi Presents: For Profit Medical Centre and get a full work up.
China tried socializing food production, and it worked terribly. Production tripled when they re-privatized it and let farmers grow for themselves. People do a better job when they get to keep the rewards.
US health care isn't capitalist in this sense of rewarding a good job. It's the worst of both worlds: a system whose regulations are superbly adapted to optimize profit for the administrative class at the expense of both doctors and patients.
Some Americans think that the US Healthcare way is “the right way but Slightly Off-Track(tm)” and will not be suaded.
You asked a direct question, though, to what extent do social services exist.
Social services exist to ensure that we all have a decent foundation on which to conduct our business of living.
For some that will be as you say, providing basic food and housing. In fact in Sweden food is given to children for free in school; however in the UK it is only poor students that get it.
In Sweden water is free, in the UK it’s charged but it’s a utility that cannot be turned off.
Everyone draws the line somewhere else, but the basis is meant to be that we have a solid foundation.
Unexpected medical expenses shouldn’t decimate a household economy for a decade, it doesn’t matter how unprepared the household is. But this is my personal feeling.
In addition: tying health insurance to employment and having at-will working conditions strikes me as ensuring compliance/docility in the work place, which I don’t believe in.
Administration bloat is a problem everywhere, however I don't think that is the reason for or against national Healthcare.
It almost certainly would be possible to use MRI for screening, but the impact would be a reduction in availability and a higher cost.
That’s absolutely fine. You’re free to spend your money on preventative healthcare that’s been determined to fail a cost benefit test. Organizations that have to save as many lives as possible or prolong healthspan as long as possible within a limited budget must make decisions somehow.
Any sort of treatment is invasive. Almost all form of medical treatment has side effects and risks.
Finding out you have "cancer" is traumatic and extremely emotional, though breast cancer is one of the most survivable (in part because, well, everyone loves boobs. Prostate cancer, on the other hand...)
Putting these tools in the hands of medical professionals is one thing. Putting them in the hands of the general public is beyond irresponsible.
People physically assaulted doctors and nurses for not being given ivermectin; imagine how insufferable people will get when some website examined their mammogram and said they have cancer.
Extremely weird take. Not least because the survival rates for breast cancer and prostate cancer are very similar.
and it's completely ridiculous why there isn't a public/government project to do this.
We need to be able to work in a world with frequent, imperfect, low cost diagnostic tools. Cancer is almost completely survivable if caught early enough. So working to figure out early detection is effectively the "cure" for cancer we have been looking for.
also it's quite possible that (just as with COVID tests) we would benefit from more testing even if that test is not that reliable. (so there's an argument for developing fast non-ionizing radiation imaging machines, eg. a fast stand-in MRI)
Anecdotally, I know (directly and indirectly) many, many women who had breast cancer before 50 years of age.
The article is a misnomer calling him an ‘amateur’. Its a click bait title. He’s shown himself to a world leading researcher in the application of AI to cancer screening.
Plenty of managers in the NHS can’t even do simple math.
See for example https://jamanetwork.com/journals/jamaoncology/fullarticle/27... for some more context.
true_pos 36
true_neg 207
false_pos 63
false_neg 16
total 322
http://mammo.neuralrad.com:5300/static/mamo.jpg
http://mammo.neuralrad.com:5300/static/mammo.jpg
(trying file names at random)
[0] https://www.repository.cam.ac.uk/handle/1810/250394
Also, you're serving up on http. Don't do that.
I think in some ways making the model available online can be good, but in other ways could be harmful too. Very complicated topic.
恭喜coolwulf， 祝你继续成功。
And especially because in the future, most radiology work will be done by software. It's just a matter of whether it's 10 years or 100 years from now.
Surely concern for the well-being of the patient figures in there somewhere...
Or imagine this: A liver lesion is incidentally discovered on your abdominal CT performed for unrelated reasons. Its radiographic characteristics are equivocal. Additional imaging studies fail to completely exclude the possibility of a liver malignancy. You undergo a biopsy. But the biopsy is complicated by hemorrhage. Surgery is required. You develop a post-operative nosocomial infection. etc. etc.
To the extent that risks along this chain of unfortunate events is known, yes, warnings could put some of the quantified decision making in the patient's hands. Well, except for the rampant innumeracy in the general population...
This topic (breast screening) is a good example due to the sheer scale. If you increase the work-up rate by even a smallish amount, you are statistically pretty much guaranteed to kill people who did not have the disease. How this balances about gain (i.e. save other lives) is not obvious. Figuring out the "right" way to do this is real work.
Gøtzsche PC. Mammography screening is harmful and should be abandoned. J R Soc Med. 2015;108(9):341-345. doi:10.1177/0141076815602452 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4582264/
It's not the case the more screening is always better. There are plenty of screening regimes that you've never heard about because the trial data didn't support it.
"I have no medical background whatsoever but I do not buy undisputed medical science."
> I'd get a CT scan every year if I could convince someone to sign off on it
20mSv of unnecessary ionizing radiation exposure a year, what could possibly go wrong? That whole-body CT scan is equivalent to getting at least EIGHTY chest x-rays, and ten times what a uranium miner receives in a year, and well within the range where an increased risk of cancer is noticeable in epidemiological data.
Case and point why this tool should not be generally available.
You would choose annual body CT even if outcomes-based evidence showed no benefit?
Even among those at highest risk for lung carcinoma, studies of annual screening chest CT do not uniformly show improved disease-specific survival as compared to an unscreened but otherwise matched cohort.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3762603/
What I found mind blowing when I first read these studies (particularly the second link) is that the positive-predictive value of the CT screen was only 3.8% for lung nodules that were 4mm or larger in size. Basically just means we find all kinds of lung abnormalities all the time that "could be cancer" but aren't.
Keep in mind, that the screening criteria was people over age 65 who had a 30+ pack-year smoking history (1 pack-year = smoking 1 pack of cigarettes per day in one year), and had to have smoked within the last 15 years.
I can also tell you from doing CT lung biopsies that it's not a procedure to take lightly. Lot of important structures in the chest.
I mean, plus I have extreme health anxiety, and I've had doctors say "nah you're too young, we don't need to test for that" and later it turned out I did have kidney failure at 35.
It really isn't. Your chances of negative consequence for unnecessary follow up procedures would rapidly become significant.
Does a patient take a negative exam as "I'm good!" and forego actually seeing a doctor, or having the exam read by a radiologist?
A positive result is also a challenging situation. I've talked to patients before after diagnosing them, and as you could expect most people are shocked, scared or they want to know what they should do next.
In breast radiology specifically (at least in the US) there is a well defined reporting lexicon and classification system. From what I can tell, the system does not use the lexicon or BI-RADS classification. https://www.acr.org/-/media/ACR/Files/RADS/BI-RADS/BIRADS-Re...
Breast cancer can be very subtle, and can be hard to figure out, especially if a patient has had surgery or other benign abnormalities. https://radiopaedia.org/cases/development-of-dcis?case_id=de...
Diagnostic and screening images also undergo rigorous QA for the entire system. Does the model have some kind of QA built in?
There is a just a small disclaimer at the top that says "we will not store your data on the server. please do not worry about privacy issues". Not exactly a formal or legal binding agreement.
In fact, you can't even connect over https. http://mammo.neuralrad.com:5300/upload
These are just a few things that come to mind off the top of my head. Like I said before, I do think that what coolwulf works on and is trying to do is good, and in the long run can help doctors better characterize findings and help patients. But like anything in health care, there are a lot of edge cases and side effects that you have to think about. The stakes are also very high.
So we don’t have to worry as much about the “Does a patient take a negative exam as ‘I'm good!’”, because in the normal flow of medicine that gets this scan the doctor will check the scan for them either way. The website probably could have better explanations of this, but most likely negatives will already be double checked independently by doctors, and positives will most likely be handled correctly by bringing it to their current current attention to double check the scan or run more tests.
I wonder how the false positive/false negative rates for this tool compare to that of a trained human. At some point we may reach parity, in which case what's the harm of trusting the automated result?
I don't think that a human reader in the loop will be going away for a long time. Sure, they can be at parity, but does a combined read have better results? This is an active research question.
I think what will happen is that over time, human interpreters and AI systems will "co-evolve" in a sense, where people will pick up on where models are wrong and also learn how to use models to understand their own blind spots. These are also active research topics that are in their infancy.
EDIT: To be clear, I believe that this is a different group of authors from coolwulf. Pubmed link:
It's our job (as radiologists and engineers) to shape that in a way that benefits patients.
This is very true. Data availability (and moreso, label availability) is the biggest barrier to improvement here I suspect [ thanks for labeling!]. Access being another. Using a public site to bootstrap that could do very interesting things. On the other hand, public access to a poorly RA/QA'd algorithm could also cause more trouble than help, easily.
I am hopeful though. Few-shot learning and self supervision are very active questions right now and there are a lot of papers in the medical AI field that are getting published on these topics.
I'm personally interested in liver cancer, which does not have large, well-curated and shared database of cases.
Sharing data gets tricky, especially in the US. Labels I'm working with and creating (at least at this point) are for my own research, which won't be shared publicly any time soon.
Most of what I've seen isn't very promising. The energy in these research areas are because it would be so much cheaper than the "right" way, far more than because of the likelihood of success. And also perversely, because it's hard for the academic researchers to get enough data to do other studies :) NB: I'm not saying there is nothing useful coming out of the learning literature in last few years, just that it a) isn't a silver bullet and b) is often being misapplied in these areas anyway.
Label quality and availability isn't the only big problem though. Many data sets exhibit problematic sampling bias, as well as being order(s) of magnitude too small, because of the way they are gathered and how access is granted.
Not to mention imaging protocols are not standardized, and the imaging technology is also evolving so scans we do today may not be "correct" or standard in 5-10 years.
I suspect we have similar overall views of the problem, but I'm pretty strongly in camp that recent advances in ML/AI are mostly really driven by data & label availability, not algorithmic advances - this colors where I think the wins to be had in medical ML can happen most easily. Either way though the non-technical barriers seem clearly higher than the technical ones still.
I’m a first year radiology registrar (PGY3) in Australia looking to find others doing interesting work in this domain, if you think I could help with your efforts feel free to DM
Results are not impressive until they are :)
It's certainly not a solved problem, and it's easy to have a pessimistic view now but I'm generally bullish on where things will be 10 years from now.
True! I certainly wouldn't discourage anyone from trying.
On the other hand, I think it would be a huge mistake to trust that fancy learning approaches will solve everything so we shouldn't try and improve access and labling. Getting better there is still by far the most high probability of successful impact, imo.
For instance, this article is silent on false positive/false negative rates of the software. There is no comparison with other research. It reads like a corporate press release promoting a product.
(for those who care- I'm a published ml biologist who works for a pharma that develops human health products. Having worked in this area for some time, I often see people who have no real idea of how the medical establishment works, or how diagnostics are marketed/sold/regulated. Overconfidence by naive individuals can have massive negative outcomes.
According to his CV he's been active in the field for quite some time. The default assumption that he's an idiot and going to kill people just seems too cynical here.
Grandparent - you specifically mention having noted methodology problems, would you mind sharing where in the methodology you think he's gone wrong?
And it doesn't matter what his credentials are, that's appeal to authority. If he thinks this should be used and trusted by people for decision making, then he should submit it to independent peer review and regulatory approval.
edit: grammar
This project might lead to people thinking they're in the clear and not seek appropriate medical treatment, or be overtreated due to an error. You should always talk to a qualified doctor if you're concerned about your health, and not use projects like these for decision making.
A false positive could create a lot of anxiety and emotional distress, and the patient might need to go to 2, 3, or 4 other doctors to get second opinions before they feel comfortable that they really don't have cancer.
A false negative could be even worse. A patient might think "oh, the official-looking online thing said I don't have cancer, so I don't need to wait for or consider a human radiologist's results", and not believe they need treatment.
I think it's very important that people understand that -- until more research is done -- this is still not a substitute for having a human look at your x-rays. If we could be reasonably sure that everyone (or at least a very large majority) understood this when using this tool, then I think people would have far fewer objections. But I don't think that's the case.
Having said that, I think it's safe to assume that this tool has saved lives, so it's almost certainly been a net positive for people.
Not even remotely the same thing.
15% of the women with breast cancer are waiting for a non-invasize diagnostic imaging system that can see their cancer. The only thing that can see these is an MRI with gadolinium. And that gadolinium contrast causes issues in about 1 in 1000 women, so it can't be used as a general screen.
I have been searching, for years, for alternative workloads for these GPUs beyond just PoW mining and password cracking. Many of them are on systems with tiny cpus, little memory, little disk, little networking so the options are heavily limited. AI/ML/Rendering/Gaming actually make bad use cases.
If anyone has thoughts on this, I'd appreciate hearing them. Let it all die is certainly an option, but it also seems just as wasteful as keeping it going. Maybe we can find a better use case, like somehow curing cancer...
Crypto markets crashing together could do this, but ETH's switch isn't going to do much for old cards.
Checking https://whattomine.com/ shows that ETH mining isn't even in the top 5 most profitable things to mine with a 1080Ti right now. The miners looking to squeeze every bit of profitability out of old hardware switched away from ETH a long time ago.
1) The cards have already paid for themselves. They are 100% ROI positive and even at the current low amounts very profitable. Regardless of what W2M says, ETH is still the top most profitable coin. Large miners don't sell immediately, they wait for the market to go up or the option against their ETH holdings.
2) ETH doesn't require latest hardware because the algo is memory hard, which means that the bottle neck is in the memory controller, not in the speed of the GPU chip itself. https://www.vijaypradeep.com/blog/2017-04-28-ethereums-memor...
3) The actual consumable is electricity price, which really hasn't changed much in the last few years for large miners who have contracts.
What? ETH isn’t the most profitable thing for these cards to mine. Why are you so confident that ETH switching to PoS will open the floodgates when ETH is already not the driver of their usage?
The ETH proof-of-stake switch isn’t going to be the driving force they causes GPUs to be abandoned in bulk.
You can mine shitcoins on a small scale for maybe a bit more profit. The issue is always unloading a lot of shitcoin in a very small volume market. Even ETC has a 6.5 day deposit time on exchanges like Kraken. Nobody is going to want to deal with that.
> The ETH proof-of-stake switch isn’t going to be the driving force they causes GPUs to be abandoned in bulk.
The GPUs will move to other shitcoins, which will cause the prices of those shitcoins to go down as people will mine and dump them, which will cause a lot of GPU miners to shut off.
The proof-of-stake switch is also pretty unlikely to ever actually happen.
1) The sum total of rewards is fixed for POW 2) Introducing extra hashing power will increase the difficulties of these mining ops up to the profitability equilibrium point.
After the overall "free" hashing power increases to a point, GPU's will start flooding the market at dumping prices.
It will be incredibly rad!
As far as I'm concerned it'll release along with Star Citizen
What we need is something that has utility... like run BOINC, earn tokens that can be used in the real world for something other than just dumping on the market.
As a consequence, it seems not worth installing on MacOS: https://stats.foldingathome.org/os
There is no incentive to run this other than good feelings. Unfortunately, that isn't enough in the business world to spend millions on cap/opx.
What I'm looking for is incentivized options. Even better if they come from a web3 situation where a business can operate without having actual customers.
"Mining", but with not such "wasteful" work.
[1] https://www.hpcwire.com/2020/10/14/how-foldinghome-identifie...
I'm just saying there is no direct financial incentive for 32 million GPUs to move over to it. If there was, they'd be on that instead of ETH.
Are you kidding? Shallow dismissals masquerading as insight are easily the most popular genre of comment on Hacker News.
> oath = "Oh you know what an alternative use is? Oaths. Works with old ASICs as well...well I think. So you take a document, like this comment, you append a nonce (you'll see) and you hash it until you get a lot of zeroes in the front. Same as bitcoin, but you're not hashing the bitcoin protocol. Then, you know the document has been sworn, as a cryptographic oath, to that extent. Nonce: 38943"
> sha256(oath) 00009ea9ab415b7f60cd43571c159d1bf1e01de4bae6a706ec9053ceb94d385c
Note the leading 0's. That's no timestamp, that's an oath.
In reply to the sibling comment: no. I like timestamp.com, and in fact I could have never found out about it other than by talking about the oath concept, but this is not just including it in the blockchain. It's proving its value to the author to bother doing the work of getting a good nonce for it. Literally putting my money where my mouth is. And swearing an oath to that extent, I could cryptographically swear it more, with more work, or use a smaller less impressive nonce if I'm not as sure.
And incentives? There is an incentive for me. At the same time it is effectively burning money, swearing by burning money. Took like seven seconds of compute, too. I had to wait human time for that. It's collateral, it's an oath. And it's an impediment to forgery, and in addition, an impediment to eg news sites telling different people different things. With oaths they have to tell everybody the same thing.
I mostly get pointed to Oauth stuff when searching for "oath sha256 nonce".
Well I suppose I can still point you to it: https://news.ycombinator.com/item?id=31451260
You know what? I'll make a post about it and link it here.
There is also very little incentive structure.
Make a token. Altruists buy the token to fund the “miners”. They may also make a profit but they buy knowing most likely they wont and it is for a good cause.
But it never took off.
The over-valuation of crypto is a two-fold negative impact on society: a massive brain-drain sucking talented engineers who would otherwise be solving real problems, and the opportunity cost of GPUs burning electricity to run unintentional ponzi schemes instead of training deep learning models.
(not that I'm a fan of crypto)
I'm not complaining about the market forces causing GPUs to be expensive, I'm complaining about the non-regulation of marketing these as securities to laymen, which is exactly what is happening right now every time you see an NFT or ICO ad on Facebook or Instagram.
Wolf on Wall Street is a movie about this happening before in history, and the current craze around crypto is exactly the same thing and the reason why the qualified investor rules were made.
If you stop dumb money being funneled into crypto, and stop the moral acceptance of chasing dumb money, you can free people to pursue more noble causes.
I'm not anti crypto, but anyone can see that a good chunk, if not a majority, of current interest in crypto is due to chasing dumb money and greed. If I recall correctly Vitalik Buterin tweeted about this excess attention before. Bad attention could kill crypto before it can truly even get its feet off the ground.
* The free AI breast cancer detection website took coolwulf about three months of spare time, sometime he had to sleep in his office to get things done, before the site finally went live in 2018
* The website also gained a lot of attention from the industry, during which many domestic and foreign medical institutions, such as Fudan University Hospital, expressed their gratitude to him by email and were willing to provide financial and technical support
* Afterwards, he and Weiguo Lu, now a tenured professor at University of Texas Southwest Medical Center, founded two software companies targeting the radiotherapy and started working on product development for cancer radiotherapy and artificial intelligence technologies
* But in 2022, he returned with an even more important "brain cancer project"
* coolwulf (Hao Jiang) (right) He told us that his parents are not medical professionals, and his interest in programming was fostered from a young age
* A reliable AI for tumor detection can enable a large number of patients who cannot seek adequate medical diagnosis in time to know the condition earlier or provide a secondary opinion
* He said that he’s not sure actually how many people have used it because the data is not saved on the server due to patient privacy concerns
Link to the technology: http://mammo.neuralrad.com:5300/
Citation Required?
The first commercially available AI/ML approach to breast cancer screening was available (US) in the late 90s. There have been many iterations and some improvements since, none of which really knock it out of the park but most clinical radiologists see the value. Perhaps the more interesting question then is why are people getting value out of uploading their own scans, i.e. why does their standard care path not already include this?
Hao
Question I have for you is that one of the biggest problems with cancer diagnoses is false positives: "Yes, there is something on your scan, we're not sure what it is, so we'll biopsy it." Biopsy is not a 0-risk procedure, and it can cause a lot of worry and pain, so it's not something to be taken lightly. Also, there are many cases of "OK, it's probably cancer, but the cure may be worse than the disease." This is the classic problem with detecting prostate cancer at an advanced age - it's very likely/probable something else will kill you before the cancer does.
How does your software deal with this issue? I'd be worried if, as you put it, people in a remote area with limited access to experienced radiologists, were given access to this and it came back with "Fairly decent chance of cancer" - what do they do then?
Hao
Even if you have done everything you are supposed to do in the process, at the end of the day you are looking at ROC curves or equivalent and trying to understand sensitivity vs. specificity trade offs, and often you have some sort of (indirect) parameters than can move to different points of that tradeoff.
This is quite critical in deployment, if you are screening you usually want something different than in diagnosis; as alluded to elsewhere if you raise the work-up rate too much you definitely risk killing more people from biopsy complications than you help with higher sensitivity (it's more complicated than that in practice)
Out of curiosity, how are you handling the data access and labelling issues here? I suspect that's the key issue that has limited the performance of the commercial offerings (hardly limited to this problem or this space).
OTOH in terms of real impact, properly leveraging a more modestly successful algorithm will probably help more people than getting a few more %. With the (strong) caveat that in a space like this you really have to look at work-up rate and balance risks.
> to at least get a second opinion on their mammogram.
for me makes a lot of sense, even in developed countries where you get a result but want extra assurances.
It would be interesting to know (assuming you have the data, even anecdotally) if the second opinions using this overturned professional ones and from those how many were corrected an original false negative mistake.
Screening breast mammo has an occurrence rate problme. Something like less than 10 in 1000 studies will require further review; this means in practice as a radiologist you look at a lot of negative films before seeing a TP. It also means a typical read is done fast. Seconds-to-small minutes.
This results in a couple of things. Reader variability based on experience/throughput, and false negatives. There were some double reader studies that caught something like 15% (going from memory here) of FN - but nobody can affort to have two radiologists read everything.
So the profession is already conceptually used to the idea of using an algorithm as a "second read" and reconsidering. Typically this won't "overturn" anything here but rather say 'hey have another look', but the decision to proceed or not is still the clinicians. Having a positive from the algorithm makes them review carefully, but you have to watch the FP rate here or nobody would get anything else done.
I have heard of health systems using algorithms as a first pass too (i.e. radiologist only see films that have had a postive in a tuned-to-be-senstive version), but that has it's own set of issues.
> borrowed from French, going back to Middle French, "one who loves, lover," borrowed from Latin amātor "lover, enthusiastic admirer, devotee," from amāre "to have affection for, love, be in love, make love to" (of uncertain origin) + -tōr-, -tor, agent suffix https://www.merriam-webster.com/dictionary/amateur#etymology...
changes the feeling of it all when you get that context, someone who loves a subject pretty much--no qualifications skill wise or regarding depth but they love it and should presumably take things seriously to some degree as any lover would.
> In 2018, a programmer named “coolwulf” started a thread about a website he had made. Users just need to upload their X-ray images, then they can let AI to carry out their own fast diagnosis of breast cancer disease.
Literally the worst fears that we have as a community is that people will recklessly apply ML to things like cancer screening on open websites and cause countless deaths, bankruptcies, needless procedures, etc. How many people went to this website, uploaded images, were told were ok and didn't follow up? How many were told they have cancer and insisted on procedures they didn't need?
The website is totally unaccountable. Totally unregulated. Totally without any of the most basic ethical standards in medicine. Without even the most basic human rights for patients. This is frankly disgusting.
In the US this would have been shut down by the FDA immediately.
We should not be celebrating this unethical "science" that doesn't meet even the most basic of scientific standards or ethical standards.
I can't believe this is getting upvoted here.
"This tool is only to provide you with the awareness of breast mammogram, not for diagnosis."
False positives are even worse, because they are far far far more likely to happen in practice. Imagine your program telling someone it has a malignant mass (VERY bad wording, only the pathologist can say something is malignant). I speak from experience that this WILL lead to the patient going to the doctor, and the doctor, seeing these very strong words, might want to take a biopsy to confirm the malignancy. These are painful procedures that can and will go wrong, eventually (complication rate is ~1%, which is not a whole lot on its own, but is unacceptable on a known healthy population). If lucky, the biopsy can be done stereotactical, but if unlucky it'll have to be done MRI-guided. You just cost society a few thousand dollars/euros/yuan. And that's if everything goes right, worst case it'll be hospital admission due to complications, like an infection.
Your blog post says you are trying to fight cancer, which is a noble cause. If the tool is not for diagnostical purposes, it's not doing a whole lot in fighting cancer as it is just a play thing then. At the moment, it's more like hindering cancer by taking resources from people who need them and giving them to people who don't need them.
Source: researcher in AI for breast cancer screening/diagnosis.
A slight digression: I find in certain countries (cough cough US), everyone is free to give advice/opinion on anything, except for medical/legal matters, which are considered sacred and so no one is free to give any advice/opinion on those. I saw a person scolded by HR for literally saying "make sure you stay well hydrated". This is madness! Medical professionals should not have a monopoly on advice concerning hydration.
Then, when you speak to the manager, ask them not to have their report telling people not to do things that HR doesn't have authority over (coworkers discussing health topics like that are not within the scope of HR).
If you are going to indirectly call someone out for publishing code which you deem severely detrimental to human health, you better back your stuff up.
> You have a false negative? Big problem
How does one get a false negative from multi-class probabilities? There aren't any...
> you just gave someone worse treatment options (if lucky) or led to someone being diagnosed too late and potentially getting metastasized breast cancer. Those bone metastases are quite painful, you know.
An image is already checked by a doctor, before given to the patient. A "False negative" then, is simply the doctor saying they are OK, and the ML confirming this, while both missing cancer. If doctor said, not OK, then no patient will take an OK from a website and be diagnosed too late. The last sentence is needlessly spiteful. I hope I misunderstood the intent of it.
> False positives are even worse, because they are far far far more likely to happen in practice. [...] I speak from experience that this WILL lead to the patient going to the doctor, and the doctor, seeing these very strong words, might want to take a biopsy to confirm the malignancy.
If doctor says OK, but my automated second opinion says not OK, then I go back to the doctor, and the doctor will either sooth my worries, explaining why the patch is benign. This is the desired outcome for my health. Or the doctor will second-guess their diagnosis, and perform, with my permission, more thorough tests. If tests come back OK, this will sooth my worries and the doctor's worries, and all will be fine. If tests come back not OK, then the application helped save my life.
There is zero reason to believe this app will bamboozle doctors into spending 1000s of dollars over nothing.
> If the tool is not for diagnostical purposes, it's not doing a whole lot in fighting cancer as it is just a play thing then.
This does not follow. There is a significant contribution with this app for fighting and raising awareness of cancer. Low-expertise hospitals use it as a diagnostics aid. People reported getting a second opinion from this app, returning to the hospital, and finding cancer early.
> At the moment, it's more like hindering cancer by taking resources from people who need them and giving them to people who don't need them.
Yes, it really seems you are deeming this app both severely dangerous and hindering cancer research. Without any solid (ML) reasoning and plenty of non-sequiturs. I'd suggest that even if you continue to hold such views, that you apologize to a fellow researcher.
Source: citizen scientist who open-sourced data analysis tools with disclaimers, now widely used for cancer screening, precision medicine, and improving chemo-therapy.
Also, 50 GPUs seems like more than necessary!
[1] https://pituitaryworldnews.org/one-worlds-greatest-neurosurg... [2] https://en.wikipedia.org/wiki/Lumbar_puncture
I will repeat what others are say: this is irresponible due to naivete and could be harmful to people. Please consult with experts on how to proceed.
Some people believe that every single person must have Mercedes-Benz type of care in the US.
They cannot fathom that some of the plebs (do they even exist for them?) may want to make their own independent healthcare choices, and are willing to accept the risk ... (or can only afford!) a Suzuki.
May not be true: https://www.washingtonpost.com/blogs/post-partisan/wp/2018/0...
The regulatory burden on medical treatments is far too high, however, and we should have the right to try: https://fee.org/archive/topics/Right%20to%20Try
What is unethical about this citizen science project? What is ethical about keeping it only for yourself, and not sharing it with the world?
You are saying you have the expertise to build a similar product, but releasing it would mean the worst fear of your community?
> people will recklessly apply ML
What are the indications that this is a reckless application of ML?
> How many people went to this website, uploaded images, were told were ok and didn't follow up?
Common sense dictates exactly zero. Their follow up was taking their images and getting an automated second opinion. Either a doctor already deemed them OK, or a doctor deemed them not OK, in which case, they would not rely on a second opinion, to think they are suddenly OK.
> How many were told they have cancer and insisted on procedures they didn't need?
Again, exactly zero. The app returns probabilities not binary diagnostics. No hospital would do anti-cancer procedures on a patient without cancer, even when they insist, because some website, friend, or religious leader told them so.
> The website is totally unaccountable.
Good. Or make the good-faith open-source project accountable and liable? That would simply mean shutting it down. No more diagnostics help for low-expertise hospitals: not good at all.
> Totally without any of the most basic ethical standards in medicine.
List a basic ethical standard in medicine which this project runs afoul of.
> basic human rights for patients
What right is that? The right not to upload your images to a site of your choosing? I thought human rights include self-determination, and keeping possession of your imaging to do however you see fit.
> In the US this would have been shut down by the FDA immediately.
But is that a good, ethical thing? Or simply that red tape and authority in US does not allow for such projects?
> We should not be celebrating this unethical "science" that doesn't meet even the most basic of scientific standards or ethical standards.
You should not talk about ethics or science, when you did not do even a proper evaluation of the work of a fellow scientist.
> I can't believe this is getting upvoted here.
Awaiting your work on cancer research and ML. Post it here. If devoid of ethical issues, and strongly scientific, it will also be upvoted and celebrated. Or is your major contribution going to be a snipe at someone who actually contributed?
is your issue with this project.. being public? not accurate enough? shouldn't be pursued at all because it'd never work?
what is the goal of regulation?
You don't get into open source development for the big bucks. It's sort of like asking someone who starts a soup kitchen why they don't open a restaurant.
I think it's worth noting how open source development spurs technical innovation in a way commercial development does not. If a company develops a proprietary technology, almost nobody will get access to the source or datasets, and so it can't then be reused and built upon to advance the state of the art. If it's open source, everyone can build on it and improve things. Sort of like how science needs to be public in order to develop more science.
But just as we need large investments to push science further, we should also have large investments in open source. We need to make access to these 50-GPU clusters easier for open source research, so a lone developer doesn't have to built it himself. Some meager funding for the actual development would be nice too... How many OSS researchers could we fund at $35K per?
Seems like 1/10 wrong would be bad, how does that compare with a doctor doing it?
On a German breast cancer screening population, 90% accuracy is abysmal as ~99.2% of cases are negative. Just predicting „no cancer“ would achieve 99.2% accuracy.
Accuracy is a very bad metric for such highly asymmetric problems.
To provide some context: The German screening system is able to identify cancer in ~6/1000 of patients screened, and missing it for the remaining 2/1000.
IIRC this is achieved by re-inviting in ~3% of cases to further examination, where ultrasound / needle biopsy / etc. can be done on a case by case basis.
> About half of the women getting annual mammograms over a 10-year period will have a false-positive finding at some point.
Any idea what the false negative is?
I guess that’s the more important factor?
In Germany, roughly 20-25% of cancers are missed in screening (and often found 2 years later)
How many really useful, cool and meaningful projects are stuck because such authors can't find or afford gpus - as they are being used to calculate meaningless hashes instead :/
In the US, the issue is getting the Xray
How is this accuracy calculated? Further in the article it is noted that there is no patient data saved by the project:
>> He said that he’s not sure actually how many people have used it because the data is not saved on the server due to patient privacy concerns. But during that time, he received a lot of thank-you emails from patients, many of them from China.
Considering user privacy is laudable in my opinion, but I'm still curious to know how accuracy is known.
I’m currently working with Digistain (S21) and we’re using AI to predict breast (and eventually other) cancer recurrence.
The tests are performed using infrared spectroscopy to measure protein synthesis and then fed into AI in order to make proper measurements and predictions.
We’ve shown we’re able to predict better than any other known method and are beginning our partnership and rollout to many hospitals around the world.
I'm a consummate algorithmist and have difficulty getting that amount hardware...for now at least. And he got them all of the same kind, that's pro. coolwulf is in fact cool.
I applaud the author. These tools seems like a great addition to health care providers. I'm just less sure about when you would use it directly as a patient.
1) Radiologists' experience on reading is not the same. The reading from a 5+ years experience radiologist sometime could be quite different from the one with a one or two years experience radiologist. As a matter of fact, due to the complicated issues with mammogram reading, some inexperienced radiologist don't get mammogram reading tasks when they started working in the clinics.
2) 2nd opinion sometime could be quite useful and important for the patients. By providing a utility for patients to have certain awareness on their mammogram, apart from the readings from their radiologists alone, could be useful, especially for people who are from remote area which lacks of experienced radiologists.
3) Even in big cities which has more medical resources (like big cities in China), due to the amount of patient, usually each radiologist only has about 30s to read a mammogram image (We have been told by multiple doctors and they could confirm that.). Mis-reading is very common due to the work-load. So a utility like this could help on finding or at lease warning doctors on possible missed.
There are other reasoning behind this but above are things on top of my head right now.
Furthermore as others have mentioned, security etc play always a bit part.
Having a doctor miss-read/miss the tumor etc can have legal actions etc. but having a machine do that is also not within the legal scope, there are so many aspects that need to be taken into consideration.
Great effort especially for the person, but science wise not much.
Also, can it tell the difference between a radiograph of a watermelon and a cancer filled breast? Prove it.
Why is Python so good? It democratizes by lowering the bar to coding.
Fsck money already, we have the power to save live beings.
I'm completely unfamiliar, but it wouldn't surprise me if for diagnosing? software like this to be used in an official medical capacity in America it would need to go through some sort of particular vetting process because if it isn't it might leave hospitals who use it open to lawsuits.
He gives a listening ear to hear you out. Greetings to you all