New Blood Test Can Detect 50 Types of Cancer (theguardian.com) 24
A new blood test that can detect more than 50 types of cancer has been revealed by researchers in the latest study to offer hope for early detection. From a report: The test is based on DNA that is shed by tumours and found circulating in the blood. More specifically, it focuses on chemical changes to this DNA, known as methylation patterns. Researchers say the test can not only tell whether someone has cancer, but can also shed light on the type of cancer they have. Dr Geoffrey Oxnard of Boston's Dana-Farber Cancer Institute, part of Harvard Medical School, said the test was now being explored in clinical trials. "You need to use a test like this in an independent group at risk of cancer to actually show that you can find the cancers, and figure out what to do about it when you find them," he said.
Writing in the journal Annals of Oncology, the team reveal how the test was developed using a machine learning algorithm -- a type of artificial intelligence. Such systems pick up on patterns within data and as a result learn to classify it. The team initially fed the system with data on methylation patterns in DNA from within blood samples taken from more than 2,800 patients, before further training it with data from 3,052 participants, 1,531 of whom had cancer and 1,521 of whom did not.
Writing in the journal Annals of Oncology, the team reveal how the test was developed using a machine learning algorithm -- a type of artificial intelligence. Such systems pick up on patterns within data and as a result learn to classify it. The team initially fed the system with data on methylation patterns in DNA from within blood samples taken from more than 2,800 patients, before further training it with data from 3,052 participants, 1,531 of whom had cancer and 1,521 of whom did not.
Cool! (Score:1)
But can it detect Covid-19? That is all that matter nowadays.
Re: (Score:2)
But can it detect Covid-19? That is all that matter nowadays.
Pretty sure people will still need treatment for, and be dying from, Cancer and other things, so this matters too -- ding dong.
Re: (Score:1)
You surely didn't read the OP nor the title itself: it reeks CANCER from each sentence and... you know what?... SARS-CoV-2 is NOT a cancer... it's, well, a SARS...
Re: (Score:2)
Being a person in a risk group of cancer, whose cancer tests have been postponed because of the Covid-19 hysteria, I am offended by your stance.
I am very afraid that I could be diagnosed with cancer too late for any treatment to work, or not at all.
Re: (Score:2)
I'm surprised and somewhat horrified that your cancer test is being postponed? Are you in an area where the hospitals are overcrowded? Or are people just trying to minimize everything to reduce spread of Covid-19?
I hope this new test is hurried into regular clinical use.
Re: (Score:2)
I've heard that a lot of hospitals are postponing all non-essential services, both to protect the general populace from their disease-ridden corridors, and to keep the beds empty for when they're needed.
Re: (Score:1)
People will still get sick. Wuflu just adds to the average hospital use.
Difficult to find a good training set (Score:5, Interesting)
There's a problem with these training sets, and that's expressed in the following part of the summary (emphasis added):
The team initially fed the system with data on methylation patterns in DNA from within blood samples taken from more than 2,800 patients, before further training it with data from 3,052 participants, 1,531 of whom had cancer and 1,521 of whom did not.
The issue is that we have ground truth for the 1,531 examples who had a diagnosed cancer. That is, we know about at least one cancer that they had. But for the 1,521 who did not have cancer, that should be re-phrased as, "1,521 who did not have a diagnosed cancer." I have seen an eminent oncologist give a talk in which he claimed that each person has, at any given time, about six cancers that the immune system is handling (this was the fellow who invented the VEGF treatment, and I'm too lazy to look his name up right now).
So the issue with the training set above is that half of the examples have a ripping cancer that is obvious enough to cause clinical symptoms. The other examples? Well, they might and they might not have a cancer. Or six. They just haven't been diagnosed.
I don't know how you get around this problem. Can anyone involved in similar work lend some insight?
Re: (Score:2)
The video, from 2013, details the lack of a theory of cancer and how the medical community had tunnel vision because of it.
Re: (Score:2)
There have been a number of theories of cancer with useful therapeutic results before Paul Davies came along. He has added an interesting additional idea to the mix, but he does fall prey to the "I'm a physicist who has figured out the key which eluded those dumb biologists for decades" habit. Some of the more interesting already-existing theories of cancer: Cancer as a Microevolutionary Process [nih.gov]. Multistage Theory [nih.gov]. Cancer is a group of more than 100 related diseases [nih.gov]. Somatic mutation theory, which Davi
Re: (Score:2)
It is difficult to believe that solving a complex problem like diagnosing a cancer, especially early, given the complexity of the factors involved will be delegated not to proper research but to bullshit like poorly-understood statistical analysis.
But yeah, we've gone there and there is no coming back - science has been reduced to Monte Carlo, "AI" or both these days.
Re: (Score:2)
It's hard to argue against results - if the poorly understood math works substantially better than any of the other options, then we have a powerful new diagnostic tool, AND the mathematical basis of that tool which we can analyze until we do understand it.
Re: (Score:3)
So the issue with the training set above is that half of the examples have a ripping cancer that is obvious enough to cause clinical symptoms. The other examples? Well, they might and they might not have a cancer. Or six. They just haven't been diagnosed.
I don't know how you get around this problem. Can anyone involved in similar work lend some insight?
The technique might or might not work. One key consideration is whether the markers for cancer gradually manifest over time (e.g., linear-like growth) or suddenly (e.g., like a step function) and coincidentally with or after the onset of symptoms observable without invasive tests. If marker growth behavior is indeed gradual and the time from the minimal threshold for AI detection to observable symptoms is significant, then this AI technique might be useful.
We'll have to wait for the double-blind tests to
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Actually you're completely off base - high false positives are a bigger issue the *lower* the number of real cases. Run the numbers (I'm doing it ough, without concerning myself about exact values)
1010 patients tested , 50% false positive, 10% false negative, 1% real cases(=10 people)
means
500 false positives, 9 real positives = testing positive means you have a 9/509 = 1.8% chance of actually having the disease - scarcely better knowledge than you had before the test
500 real negatives, 1 false negative =
Re: (Score:2)
56% false positive is only bad if the rate of actual is very high.
I think there must be some confusion here, it's a 56% false negative rate, not false positive. It tells you that you don't have cancer, but actually you do.
Re: (Score:2)
After reading more of the article, it sounds more like the 56% false negative rate. They have poor wording. I now agree wi
Re: (Score:2)
There is also another pretty bad problem: In actual reality, the situation is not that half of the population has cancer. Unless your ground truth training set actually represents reality, you will get very skewed results.
hmm (Score:2)
As a cancer survivor, I would love this tech to be effective and rolled out quickly. However, we have literally seen this exact story ("Scientists Can Detect Cancer With a Simple Blood/Urine/Whatever Test!") for decades and they never seem to actually be used.
Re: (Score:2)
Well, they probably used synergized cloud-based AI over a fibre-rich network to provide real-time optimized detections this time.
Re: (Score:2)
Unfortunately, even an extremely effective diagnostic test normally needs years of expensive FDA testing for safety and efficacy. And the medical companies quite often have more profitable things in the pipeline to move forward. Especially if they're already selling an existing test whose sales would be cannibalized by the new one, there's just not much benefit for them to get the new one approved unless one of their competitors is working on getting approval for a similarly improved option.
They don't car