Medical Diagnostic Imaging, Leukemia, and Black Holes

post by nixtaken · 2020-08-03T05:40:41.128Z · score: -11 (5 votes) · LW · GW · 4 comments

Medical diagnostic imaging is highly dependent on the operator because whenever a radiologist looks at a set of MRI images or CAT scans, he or she will adjust the contrast and select a handful of images to tell a story about what he thinks is going on inside of a patient. Often this story is influenced by what the doctor who prescribed the MRI has written and more than ever, the story filtered out is guided by automated or AI tools.

Since we often see the things we are told to see, doctors have learned to be careful about diagnosis and make sure that they use several, independent sources of evidence.

If you only use a single diagnostic device for a disease, it is very easy to end up with a false-positive result and see a disease where none exists. For example, a doctor might mistake a blurry spot on the image for cancer or he might filter an image based on a diagnosis that he wants to see.

False positives are reduced by using multiple, independent diagnostic machines and blood tests. If they all agree with the symptoms reported by the patient and if you are lucky enough that those who performed the tests were not influenced by what they thought they should measure or what they were afraid might be measured by someone else (making them look bad), then when the surgeon shows up, he won’t cut you open and find nothing to remove.

If medical professionals don’t understand the necessity of having multiple independent, unbiased measurements, people die, but this is not the case in the field of astronomy and physics.


I cross-posted these articles here on Less Wrong because I think it is important for computer scientists and data scientists to be aware of some of the irrationality infecting portions of the physics community. This is important because ex-physicists are frequently embraced by the computer science and data science communities and they are entrusted with complex, critical tasks, based on assumptions about their understanding of how to handle data. They may end up building tools that make medical diagnoses.

In these articles, I wrote about some egregious examples of physicists handling data poorly, but within a few days, the welcoming committee here gave them a Karma of -89. I was hoping to find readers who might make use of this information, but so far, I've met with surprising forms of resistance.

I was rather appalled to see a seemingly intelligent LessWrong commenter argue that

If I didn’t know better, I would think that this gish galloper was acting like a caricature of irrationality and executing a form of psychological warfare designed to discourage people like me from criticizing certain sacred cows of the scientific community. A more charitable assessment is that my articles provoked a strong case of cognitive dissonance in him. The moderator not only supported this commenter's logic but he also removed the scientific methods and philosophy tag, even though that was the main topic of the posts. My hope is that I just attracted a few bad apples and that other people will understand why I've come here.


If an EHT black hole astronomer combines blurry images from a bunch of different telescopes in order to create a picture of something that he believes should exist in unimaginably distant space, nobody is going to die.. at least not immediately.

If a LIGO gravitational waves astronomer filters out a blip buried in noise and attributes it to black holes colliding in a distant star system instead of to something more mundane and terrestrial in origin, nobody is going to die.. at least not immediately.

We all die eventually, but some of us may realize that we’ve wasted our lives measuring things that do not exist. We also might realize that a corruption of the scientific method within physics might infect other, more critical disciplines, like medicine. Many ex-physicists go on to do data analysis for medical research or banking.

This sort of confusion arises when astronomers believe that:

The first point confuses people because they might think about how a pendulum oscillates randomly about a central point and how if you average over enough measurements, you will get a better measurement of that point relative to the fixed point holding the pendulum. If you take enough measurements, the difference between the relative positions of the fixed point and the measured point should go to zero – in an ideal, classical world.

In the real world, one might imagine the fixed point and the measured point wobbling around relative to everything else, as though they were sitting on a boat in the sea. The measured point and the fixed point may always be connected by a straight line, but that line may have no fixed relationship to anything else. In quantum mechanics, this is called intrinsic uncertainty, but in astronomy, people act as though this uncertainty doesn’t exist, because it doesn’t exist in general relativity.

In the real world, the Earth is moving relative to the star you want to measure and you cannot anticipate all of those movements or distinguish them from how light bends as it travels between the star and the Earth. Even if you have perfect equations governing the motion of an object, without perfect knowledge of initial conditions, there will always be some uncertainty about the predicted motion. Nothing is truly stationary relative to anything else and this motion will define a noise floor below which you simply can’t measure anything.

Far too many experiments in physics today are attempting to measure things that are below the noise floor and they are doing this without using any of the basic principles of the scientific method.

Strange things happen when medical professionals try to measure things that are below the noise floor of their measurement devices. They may try to combine measurements to see things that cannot be seen with an individual measurement and sometimes this appears to benefit the patient, but sometimes it does not. It is possible for a group of measurements that are all below the noise floor to give false positives that lead to a positive diagnosis — just to err on the side of caution in case of future malpractice lawsuits.

Of course, if you understand the limitations imposed by a noise floor, using multiple measurement methods to cross-check a hypothesis is clearly better than relying on a single measurement device. But even then, people can make mistakes when they combine uncertain, biased measurements and overestimate their certainty.

That was what I was thinking as I read this story about a perfectly healthy man who was diagnosed with leukemia and then blasted full of chemo, leading to years of trauma and recovery from something that was probably never going to become a problem.

I looked up the diagnostic criteria for leukemia and found that it is not a disease that you can catch in its early stages through blood tests. It has to present through symptoms – night sweats, fevers, infections, tiredness, weakness, bruising, paleness, bleeding, etc. He had none of these symptoms. Recall that before all of this started, this guy was as healthy as anyone could be, yet he was diagnosed with a leukemia precursor (MRD) based on a blood test that produces false positive results. A series of doctors who didn’t want to miss a diagnosis gave him a diagnosis (MRD that might become leukemia) because the penalties for missing a leukemia diagnosis were far worse for them than for over-diagnosing and treating leukemia. They were protecting themselves rather than the patient.

The analysis of a sample is very complicated and unreliable. I bet a physicist designed it.

I am not a doctor and I hope that nobody takes my advice in their own medical care or in the care of their child, but I detect a systematic problem here and the problem is the over-reliance on tests that are not as reliable as those who use them believe. I mean, if you or your child are clearly sick, you should get help from a doctor, but this man’s story bothered me because he was perfectly healthy until the doctors told him that he was sick and gave him an invasive cancer treatment.

This faith in the medical system is quite similar to the faith people have in measurements of gravitational waves or the faith they have in the FDA and in dermatologist’s recommendations to use sunscreen.

What if it is all rotten to the core and we should be working harder to figure out which parts of the system can be salvaged. Do we need to turn back the clock and start doing science and medicine in the same way it was done back in the 1950s? Something went wrong in the past 70 years and I think it was our overzealous faith in technology.

Caveat Emptor! Experts are often stupid. So what are you going to do? Some people want to create AI tools to help individual patients evaluate their own situation, rather than relying on a bunch of doctors who might possibly be idiots or corrupted by their incentives.

Some of our founding inspirations were the work of Gerd Gigerenzer and Atul Gawande, who showed that medical errors were the cause of many deaths, that doctors tend to be statistically illiterate, and that systematizing tools like checklists and statistical prediction rules save lives. [LW · GW]

I do feel sorry for the medical professionals who are fighting the coronavirus and I wish them the best for themselves and their families, but I do not feel sorry for the medical industry as a whole. Just like the physics community, it needs a hard reset and re-grounding in foundational principles.

When I see experiments like LIGO and EHT given top billing in the science media, I see this as a symptom of a much wider problem with people mistaking faith in science for an understanding of the scientific method.

Consider that all of these physics experiments are used as training facilities for the people who go on to interpret data from fields as diverse as agriculture, economics, and medicine.

I think we need to fix this widespread educational problem before ex-physicist quants create another banking crisis or before a miseducated ex-physicist convinces a doctor to use his software package to make decisions about a kid’s leukemia diagnosis.


Comments sorted by top scores.

comment by Viliam · 2020-08-03T15:41:31.968Z · score: 28 (8 votes) · LW(p) · GW(p)

Uhm. Not sure how to put this, but seems like remaining silent would be a worse option...

Looking at your other articles, it seems to me like you came here with many strong opinions you want to share. You keep posting long monologues, and you get some response to them, but I suspect that there are more articles already written and waiting to be published, regardless of the feedback you might get here.

Which wouldn't be a problem if your articles were positively accepted (i.e. upvoted) by the community... but they are not. Barely anyone votes on them, and the average feedback is negative.

I wish I could provide a more detailed feedback, but the fact is that these articles are just too long for me. Even length is not exactly the problem, it's like... for example, this article is about medical diagnostic imaging, which is a topic I know almost nothing about, your arguments are not easy for me to verify, maybe you are right about something important, or maybe you are completely wrong... I can't tell. Generally, if I can't tell, I don't vote. But I suspect that most readers here are in the same position. But then, this is not the right audience for you, and it is not the right content for this website.

If you want actual engagement with your ideas, I suggest to take it much slower. Don't put dozen claims in a single article, because almost no one will read them all. (Ironically, you mention "gish gallop" in the same article than contains links to proteins, group delusion, lithium, gravitational waves, leukemia diagnostics, sunscreen, Triclosan, spine surgery, Gadolinium, boron, flouride, heart attack, neutrinos, Higgs particle, satellites, balloons, laser, fusion, climate, planets, and expansion of the universe.) If you focus on one thing, there is a chance you may find people who understand it and can have a debate with you. Otherwise, you are just wasting time here.

comment by nixtaken · 2020-08-03T17:12:10.027Z · score: -12 (8 votes) · LW(p) · GW(p)

I see that of your 28 posts, most of them are related to organizing Less Wrong meetups. That must be a great way to build up Karma. You are a community builder and I imagine that your instinct is to avoid confrontation, make friends with the right people, and support group norms. This is essential if you want a community to grow, but if you want a community to maintain a high standard of inquiry and rationality, controversy and ideas that are not universally popular must be tolerated.

You seem to dismiss this post based on the wide range of physics topics to which it links, but it does have a clear narrative and focus: how educational corruption in physics might affect something as seemingly unrelated as medicine.

comment by Daniel Kokotajlo (daniel-kokotajlo) · 2020-08-03T18:41:18.741Z · score: 12 (5 votes) · LW(p) · GW(p)

Just chiming in to say I agree with Villiam. I actually read your whole post just now, and I thought it was interesting, and making an important claim... but I couldn't follow it well enough to evaluate whether or not it is true, or a good argument, so I didn't vote on it. I like the suggestion to break up your stuff into smaller chunks, and maybe add more explanation to them also.

For example, you could have made your first post something like "Remember that famous image of a black hole? Guess what: It may have been just hallucinating signal out of noise. Here's why." Then your next post could be: "Here's a list of examples of this sort of thing happening again and again in physics, along with my general theory of what's wrong with the epistemic culture of physics." I think for my part at least, I didn't have enough expertise to evaluate your claims about whether these examples really were false positives, nor enough expertise to evaluate whether they were just cherry-picked failures or indicative of a deeper problem in the community. If you had walked through the arguments in more detail, and maybe explained some key terms like "noise floor" etc., then I wouldn't have needed expertise and would be able to engage more productively.

comment by lsusr · 2020-08-04T20:42:55.665Z · score: 7 (5 votes) · LW(p) · GW(p)

I myself write controversial posts like this [LW · GW] which often combine ideas from a broad range of disciplines [LW · GW]. I have even written specifically about problems in theoretical physics [? · GW]. As something of a heretic [LW · GW], I appreciate Villiam's work as a community builder and I agree with every one of Villiam's points.

This community recognizes that there are indeed problems in the application of statistical reasoning within physics and biology. I suspect the strongest negative reaction of this community towards you has little to do with your opinions. You are getting a negative reaction for your poor writing quality.

Your posts are full of factual errors. [LW(p) · GW(p)] They are full of logical errors. [LW(p) · GW(p)] They contain misleading information [LW(p) · GW(p)]. They are written in a disdainful tone inappropriate for this site. Such writing does not meet the criteria of "a high standard of inquiry".

Worst of all, you insult community members.

…a seemingly intelligent LessWrong commenter…

…A more charitable assessment is that my articles provoked a strong case of cognitive dissonance in him…

That must be a great way to build up Karma.…I imagine that your [Villiam's] instinct is to avoid confrontation, make friends with the right people, and support group norms…

Please refrain from insulting other community members. That is not how we do things here.