this headline is a bit misleading on the first read, since it only affects functional (f)MRI, which is controversial since a longer time. a prominent example is the activity that has been detected in a dead salmon
If you apply enough gain and filtering to an unknown signal, eventually you'll pull something out of it that you can convince yourself is what you're looking for.
Why did TUM let this misleading headline front the news release? Dont we have enough issues with Academia? The result just mean BOLD is an imperfect proxy.
As someone who used to work at the Cognitive Neurophysiology Lab in the Scripts Institute-- doing some work on functional brain image-- I can confirm this was not news even thirty years ago. I guess this is trying to make some point to lay people?
This isn’t entirely news to people in the field doing research, but it’s important information to keep in mind when anyone starts pushing fMRI (or SPECT) scans into popular media discussions about neurology or psychiatry.
There have been some high profile influencer doctors pushing brain imaging scans as diagnostic tools for years. Dr. Amen is one of the worst offenders with his clinics that charge thousands of dollars for SPECT scans (not the same as the fMRI in this paper but with similar interpretation issues) on patients. Insurance won’t cover them because there’s no scientific basis for using them in diagnosing or treating ADHD or chronic pain, but his clinics will push them on patients. Seeing an image of their brain with some colors overlayed and having someone confidently read it like tea leaves is highly convincing to people who want answers. Dr. Amen has made the rounds on Dr. Phil and other outlets, as well as amassing millions of followers on social media.
Pop science guru-ing is a giant flashing red sign for me. I am never even a little surprised when the latest “sense maker” or pop science guru comes out as a complete loon or is consumed by some kind of scandal.
Influencers in general are always suspect. The things that get you an audience fast are trolling or tabloid-ish tactics like conspiracism.
There are good ones but you have to be discerning.
I saw a clinical report of his on a patient, he puts a graphic in their report of their "brain scan" but it's basically a vector graphic of the brain w/ a multicolor MS Paint gradient...
>> Seeing an image of their brain with some colors overlayed ... is highly convincing
Indeed, there's been quite a few studies [1] that find just including any old image of a brain with stuff highlighted will cause a paper to be perceived as more scientifically credible.
This study is validating a commonplace fMRI measure (change in blood-oxygenation-level-dependent or BOLD signal) by comparing it with a different MRI technique, one that uses a multiparametric quantitative BOLD model, a different model for BOLD derived from two separate MRI scans which measure two different kinds of signal (transverse relaxation rates), and then multiply/divide by a bunch of constants to get at a value.
I'm a software engineer in this field, and this is my layman-learns-a-bit-of-shop-talk understanding of it. Both of these techniques involve multiple layers of statistical assumptions, and multiple steps of "analysing" data, which in itself involves implicit assumptions, rules of thumb and other steps that have never sat well with me. A very basic example of this kind of multi-step data massaging is "does this signal look a bit rough? No worries, let's Gaussian-filter it".
A lot of my skepticism is due to ignorance, no doubt, and I'd probably be braver in making general claims from the image I get in the end if I was more educated in the actual biophysics of it. But my main point is that it is not at all obvious that you can simply claim "signal B shows that signal A doesn't correspond to actual brain activity", when it is quite arguable whether signal B really does measure the ground truth, or whether it is simply prone to different modelling errors.
In the paper itself, the authors say that it is limited by methodology, but because they don't have the device to get an independent measure of brain activation, they use quantitative MRI. They also say it's because of radiation exposure and blah blah, but the real reason is their uni can't afford a PET scanner for them to use.
"The gold standard for CBF and CMRO2 measurements is 15O PET; but this technique requires an on-site cyclotron, a sophisticated imaging setup and substantial experience in handling three different radiotracers (CBF, 15O-water; CBV, 15O-CO; OEF, 15O-gas) of short half-lives8,35. Furthermore, this invasive method poses certain risks to participants owing to the exposure to radioactivity and arterial sampling."
This is why I love this site. You get input from so many specialized folks! I appreciate you contributing your expertise and I also appreciate you calling out the limits to that knowledge.
Two points I'm hoping you can help clarify:
> Researchers ... found that an increased fMRI signal is associated with reduced brain activity in around 40 percent of cases.
So it's not just that they found it was uncorrelated, they found it was anticorrelated in 40% of cases?
And you are suggesting that conclusion suffers from the same potential issues as these fMRI studies in general?
Like you mention, it seems to me if we wanted to really validate the model, we'd have to run the same experiment with two, three, or maybe even more different modalities (fMRI, PET with different tracers, etc).
My previous job was at a startup doing BMI, for research. For the first time I had the chance to work with expensive neural signal measurement tools (mainly EEG for us, but some teams used fMRI). and quickly did I learn how absolute horrible the signal to noise ratio (SNR) was in this field.
And how it was almost impossible to reproduce many published and well cited result. It was both exciting and jarring to talk with the neuroscientist, because they ofc knew about this and knew how to read the papers but the one doing more funding/business side ofc didn't really spend much time putting emphasis on that.
One of the team presented a accepted paper that basically used Deep Learning (Attention) to predict images that a person was thinking of, from the fMRI signals. When I asked "but DL is proven to be able to find pattern even in random noise, so how can you be sure this is not just overfitting to artefact?" and there wasn't really any answer to that (or rather the publication didn't take that in to account, although that can be experimentally determined). Still, a month later I saw tech explore or some tech news writing an article about it, something like "AI can now read your brain" and the 1984 implications yada yada.
So this is indeed something probably most practitioners, masters and PhD, realize relatively early.
So now that someone says "you know mindfulness is proven to change your brainwaves?" I always add my story "yes, but the study was done with EEG, so I don't trust the scientific backing of it" (but anecdotally, it helps me)
There's fancier ML studies on EEG signal but probably not consistent enough for clinical work. For now, the one thing EEG can reliably tell is if you're having a seizure or not, if you're delirious (or in a coma) or not, or if you're asleep.
> When I asked "but DL is proven to be able to find pattern even in random noise, so how can you be sure this is not just overfitting to artefact?"
So here you say quite a mouthful. If you train it on a pattern it'll see that pattern everywhere - think about the early "Deep Dream" trippy-dogs-pictures nonsense that was pervasive about eight or nine years ago.
I repaired a couple of cameras for someone who was working with a large university hospital about 15 years ago, where they were using admittedly 2010s-era "Deep Learning" to analyse biopsy scans for signs of cancer. It worked brilliantly, at least with the training materials, incredible hit rate, not too terrible false positive rate (no biggie, you're just trying to decide if you want to investigate further), really low false negative rate (if there was cancer it would spot it, for sure, and you don't want to miss that).
But in real-world patient data it went completely mental. The sample data was real-world patient data, too, but on "uncontrolled" patients, it was detecting cancer all over the place. It also detected cancer in pictures of the Oncology department lino floor, it detected cancer in a picture of a guy's ID badge, it detected cancer in a closeup of my car tyre, and it detected cancer in a photo of a grey overcast sky.
Aw no. Now what?
Well, that's why I looked at the camera for them. They'd photographed the biopsies with one camera on site, from "real patients", but a lot of the "clear" biopsies were from other sites.
You're ahead of me now, aren't you?
The "Deep Learning" system had in fact trained itself on a speck of shit on the sensor of one of the cameras, the one used for most of the "has cancer" biopsies and most of the "real patient under test" biopsies. If that little blob of about a dozen slightly darker pixels was present, then it must be cancer because that's what the grown-ups told it. The actual picture content was largely irrelevant because the blob was consistent across all of them.
I'm not too keen on AI in healthcare, not as a definitive "go/no-go" test thing.
I'm not sure I understand. Wouldn't any prediction result above statistical random (in the image mind reading study) be significant? If the study was performed correctly I don't really need to know much about fMRI to tell whether it's an interesting result or not.
90% of papers I read in computer science / computer security speak of software written or AI models they trained that are nowhere to be found. Not on git nor via email to the authors.
Saw the same thing first hand with Pathology data. Image analysis is far more straightforward problem than fMRI, but sorry, I do not trust your AI model that matches our pathologist’s scoring with 98.5% accuracy. Our pathologists are literally guesstimating these numbers and can vary by like 10-20% just based on the phase of the moon, whether the pathologist ate lunch yet, what slides he looked at earlier that day…that’s not even accounting for inter-pathologist variation…
Also saw this irl with a particular NGS diagnostic. This model was initially 99% accurate, P.I. smelled BS, had the grad student crunch the numbers again, 96% accurate, published it, built a company around this product —-> boom, 2 years later it was retracted because the data was a lot of amplified noise, spurious hits, overfitting.
I don’t know jack compared to the average HN contributor, but even I can smell the BS from a mile away in some of these biomedical AI models. Peer review is broken for highly-interdisciplinary research like this.
I think you're throwing the baby out with the bathwater, while also pointing to the missing pieces in our understanding of the brain and consciousness.
I also work in the field, specifically with sleep slow-wave enhancement.
Blood flow as a proxy for brain activity I always felt was a weak measure, as the brain activity involved across all manner of operating our biological systems, so is the increased blood flow measured in fMRI a response to cognition, or autonomic activity? What does that oxydation mean.
EEG is similarly flawed when we try to equate "brainwaves" to emotions and consciousness. I think we're almost better off measuring HRV, a much simpler measure, and more reliable.
I'm fascinated that so many people who discuss brainwaves think of them as actual "waves", when it is just how we plot electrical activity which creates a visual wave like pattern.
However, and this is specifically related to our work in sleep, we can detect slow-waves (I dislike that term, it's the synchronous firing of neurons) and we are able to stimulate this restorative brain function through sensory perception during sleep, and even create slow-waves in a lab using TMS.
Research linked on our website [1]
I agree the industry needs to stop projecting what we hope we're seeing with what is actually being measured, and we don't understand enough about how the brain works, but I think completely throwing away any brain related measures we have is going too far.
For task fMRI, the test-retest reliability is so poor it should probably be considered useless or bordering on pseudoscience, except for in some very limited cases like activation of the visual and/or auditory and/or motor cortex with certain kinds of clear stimuli. For resting-state fMRI (rs-fMRI), the reliabilities are a bit better, but also still generally extremely poor [1-3].
There are also two IMO major and devastating theoretical concerns re fMRI that IMO make the whole thing border on nonsense. One is the assumed relation between the BOLD signal and "activation", and two is the extremely horrible temporal resolution of fMRI.
It is typically assumed that the BOLD response (increased oxygen uptake) (1) corresponds to greater metabolic activity, and (2) increased metabolic activity corresponds to "activation" of those tissues. This trades dubiously on the meaning of "activation", often assuming "activation = excitatory", when we know in fact much metabolic activity is inhibitory. fMRI cannot distinguish between these things.
There are other deeper issues, in that it is not even clear to what extent the BOLD signal is from neurons at all (could be glia), and it is possible the BOLD signal must be interpreted differently in different brain regions, and that the usual analyses looking for a "spike" in BOLD activity are basically nonsense, since BOLD activity isn't even related to this at all, but rather the local field potential, instead. All this is reviewed in [4].
Re: temporal resolution, essentially, if you pay attention to what is going on in your mind, you know that a LOT of thought can happen in just 0.5 seconds (think of when you have a flash of insight that unifies a bunch of ideas). Or think of how quickly processing must be happening in order for us to process a movie or animation sequence where there are up to e.g. 10 cuts / shots within a single second. There is also just biological evidence that neurons take only milliseconds to spike, and that a sequence of spikes (well under 100ms) can convey meaningful information.
However, the lowest temporal resolutions (repetition times) in fMRI are only around 0.7 seconds. IMO this means that the ONLY way to analyze fMRI that makes sense is to see it as an emergent phenomenon that may be correlated with certain kinds of long-term activity reflecting cyclical BOLD patterns / low-frequency patterns of the BOLD response. I.e. rs-fMRI is the only fMRI that has ever made much sense a priori. The solution to this is maybe to combine EEG (extremely high temporal resolution, clear use in monitoring realtime brain changes like meditative states and in biofeedback training) with fMRI, as in e.g. [5]. But, it may still well be just the case fMRI remains mostly useless.
[1] Elliott, M. L., Knodt, A. R., Ireland, D., Morris, M. L., Poulton, R., Ramrakha, S., Sison, M. L., Moffitt, T. E., Caspi, A., & Hariri, A. R. (2020). What Is the Test-Retest Reliability of Common Task-Functional MRI Measures? New Empirical Evidence and a Meta-Analysis. Psychological Science, 31(7), 792–806. https://doi.org/10.1177/0956797620916786
[2] Herting, M. M., Gautam, P., Chen, Z., Mezher, A., & Vetter, N. C. (2018). Test-retest reliability of longitudinal task-based fMRI: Implications for developmental studies. Developmental Cognitive Neuroscience, 33, 17–26. https://doi.org/10.1016/j.dcn.2017.07.001
[3] Termenon, M., Jaillard, A., Delon-Martin, C., & Achard, S. (2016). Reliability of graph analysis of resting state fMRI using test-retest dataset from the Human Connectome Project. NeuroImage, 142, 172–187. https://doi.org/...
re: your last point that is not true. we can measure arbitrarially quickly (Nottingham group does some 3d EVI at ~100ms TRs). You can also reduce volumes and just look at single slices etc, a lot of the fundamental research did this (wash U / Minnesota / etc in the 90s). Its just not all that useful because the SNR tanks and the underlying neurovascular response is inherently low-pass. There is a much faster 'initial-dip' where the bold signal swings the other way and crosses zero (from localized accumulation of DeoxyHg before the inrush of OxyHg from the vascular response). Its a lot better correlated with LFP / spiking measures but just very hard to measure on non-research scanners...
I wonder how much variation there is between a person who does certain mental activity regularly vs a person who rarely does it.
If they were to measure a person who performs mental arithmetic on a daily basis, I'd expect his brain activity and oxygen consumption to be lower than those of a person who never does it. How much difference would that make?
I worked in an fMRI lab briefly as a grad student. I suspect you'd be correct but perhaps not exactly why you'd expect. Studies using fMRI measure a blood-oxygenation-level-dependent (BOLD) signal in the brain. This is thought to be an indirect measure of neural activity because a local increase in neural firing rate produces a local increase in the need for, and delivery of, oxygenated blood.
The question then is, do you expect a person who is really good at mental arithmetic to have less neural firing on arithmetic tasks (e.g., what is 147 x 38) than the average joe. I would hypothesize yes overall to solve each question; however, I'd also hypothesize the momentary max intensity of the expert to peak higher. Think of a bodybuilder vs. a SWE bench-pressing 100 lbs for 50 reps. The bodybuilder has way more muscle to devote to a single rep, and will likely finish the set in 20 seconds, while the SWE is going to take like 30 minutes ;)
fMRI is a cool, expensive tech, like so many others in genetics and other diagnostics. These technologies create good jobs ("doing well by doing good").
But as other comments point out, and practitioners know, their usefulness for patients is more dubious.
In related news: ironically, Psychedelics disrupt normal link between brain’s neuronal activity and blood flow - thus casting some doubt on findings that under psychedelics more of the brain is connected (since fMRI showed elevated blood flow, suggesting higher brain activity).
I'll get raked for this, but as someone in the field, I can say with high confidence that the majority of comments in this thread are not from imaging experts, and mostly (mis)informed by popular science articles. I do not have the time to properly respond to each issue I see. The literature is out there in any case.
> I do not have the time to properly respond to each issue I see. The literature is out there in any case.
I think your expertise would be very welcome, but this comment is entirely unhelpful as-is. Saying there are bad comments in this thread and also that there is good literature out there without providing any specifics at all is just noise.
You don't have to respond to every comment you see to contribute to the discussion. At minimum, could you provide a hint for some literature you suggest reading?
There are literally less than 20 top level comments and this one is (at least for me) the 2nd or 3rd.
Instead of a nothingburger, you could have used your academic prowess to break down the top 1/2 misconceptions with expertise.
You might not have time to respond to all the comments but a couple of clarifications could have helped anyone else who doesn't comment without experience.
Just saying that next time you can be the change you want to see in HN instead of wasting text telling us how ignorant we are.
I'm not a specialist by any means, although I am a patient of an fMRI. One thing I will note is that in the eventual, resultant paperwork from the broad array of tests I had, the fMRI was not noted whatsoever, neither was it discussed with me by any of the numerous neurologists or surgeons involved in my case. I was quite curious as to why it was performed at all, but presumably it was some formality to check a box.
agree. especially the comments saying "just address it". Its a lot of technically complicated interactions between the physics, imaging parameters, and processing techniques.
Unfortunately the end users (typically neuroscience/psych grad students in labs with minimal oversight) usually run studies that just "throw everything at the wall and see what sticks" not realizing that is the antithesis of the scientific method. No one goes in to a resting state study saying "we're going to test if the resting state signal in the <region> is <changed somehow> becuase of <underlying physiology>".
They instead measure a bunch of stuff find some regions that pass threshold in a group difference and publish it as "neural correlates of X". Its not science, and its why its not reproducible. People have build whole research programs on noise.
I remember reading a paper back in grad school where the researchers put a dead salmon in the magnet and got statistically significant brain activity readings using whatever the analysis method à la mode was. It felt like a great candidate for the Ig Nobel awards.
If you actually read their paper, you will find that it's only the sign of the correlation that is being questioned. The field has generally been aware of this interpretational gap, and that's why two-sided hypothesis tests are important. Cellular neuroscience and electrophysiology are only starting to face the challenges that fMRI faced 2 decades ago.
To me this is like shitting on cars in 1925 because they kill people every now and then. Cars didn't go away, and nor will fMRI, until someone finds a better way to measure living people's brains.
TUM's press is being sloppy, from conflating fMRI with MRI to presuming this is revolutionary, and ignoring earlier empirical work against this narrative (Windkessel's, Logothetis beta/gamma coupling, etc.)
What's surprising is the desire to have a silver bullet, or not solution.
What's still amazing is fMRI can provide more visual context of what's happening in the brain, in what region, and activities that can help that improve.
There are other complementary technologies like QEEG and SPECT that can also shed a light as well.
It does seem the case that fMRI cann be more of a snapshot photo, and technologies like SPECT can provide more of a regional time lapse of activity.
I was a grad student at UCSD when Ed Vul published Voodoo Correlations in Social Neuroscience [1], which stoked a severe backlash from the fMRI syndicate resulting in a title change to Puzzlingly High Correlations in fMRI Studies of Emotion, Personality, and Social Cognition [2]. There is a lot of interesting commentary around this article (e.g., “Voodoo” Science in Neuroimaging: How a Controversy Transformed into a Crisis [3]). To me it was fascinating to watch Vul (an incredibly rare talent, perhaps a genius), take on an entire field during his 1st year as assistant professor.
I might be oversimplifying, but isn't a lot of our neurological understanding about ADHD based on "fMRI shows decreased activity in the frontal cortex"? Or for that matter, our neurological understanding of a lot of mental health conditions.
I know the actual diagnosis is several times more layered than this attempt at an explanation, but I always felt that trying to explain the brain by peering at it from outwards is like trying to debug code by looking at a motherboard through a bad microscope.
54 comments
[ 4.1 ms ] story [ 65.0 ms ] threadThere have been some high profile influencer doctors pushing brain imaging scans as diagnostic tools for years. Dr. Amen is one of the worst offenders with his clinics that charge thousands of dollars for SPECT scans (not the same as the fMRI in this paper but with similar interpretation issues) on patients. Insurance won’t cover them because there’s no scientific basis for using them in diagnosing or treating ADHD or chronic pain, but his clinics will push them on patients. Seeing an image of their brain with some colors overlayed and having someone confidently read it like tea leaves is highly convincing to people who want answers. Dr. Amen has made the rounds on Dr. Phil and other outlets, as well as amassing millions of followers on social media.
Influencers in general are always suspect. The things that get you an audience fast are trolling or tabloid-ish tactics like conspiracism.
There are good ones but you have to be discerning.
Indeed, there's been quite a few studies [1] that find just including any old image of a brain with stuff highlighted will cause a paper to be perceived as more scientifically credible.
[1] https://pubmed.ncbi.nlm.nih.gov/17803985/
I'm a software engineer in this field, and this is my layman-learns-a-bit-of-shop-talk understanding of it. Both of these techniques involve multiple layers of statistical assumptions, and multiple steps of "analysing" data, which in itself involves implicit assumptions, rules of thumb and other steps that have never sat well with me. A very basic example of this kind of multi-step data massaging is "does this signal look a bit rough? No worries, let's Gaussian-filter it".
A lot of my skepticism is due to ignorance, no doubt, and I'd probably be braver in making general claims from the image I get in the end if I was more educated in the actual biophysics of it. But my main point is that it is not at all obvious that you can simply claim "signal B shows that signal A doesn't correspond to actual brain activity", when it is quite arguable whether signal B really does measure the ground truth, or whether it is simply prone to different modelling errors.
In the paper itself, the authors say that it is limited by methodology, but because they don't have the device to get an independent measure of brain activation, they use quantitative MRI. They also say it's because of radiation exposure and blah blah, but the real reason is their uni can't afford a PET scanner for them to use.
"The gold standard for CBF and CMRO2 measurements is 15O PET; but this technique requires an on-site cyclotron, a sophisticated imaging setup and substantial experience in handling three different radiotracers (CBF, 15O-water; CBV, 15O-CO; OEF, 15O-gas) of short half-lives8,35. Furthermore, this invasive method poses certain risks to participants owing to the exposure to radioactivity and arterial sampling."
Two points I'm hoping you can help clarify:
> Researchers ... found that an increased fMRI signal is associated with reduced brain activity in around 40 percent of cases.
So it's not just that they found it was uncorrelated, they found it was anticorrelated in 40% of cases?
And you are suggesting that conclusion suffers from the same potential issues as these fMRI studies in general?
Like you mention, it seems to me if we wanted to really validate the model, we'd have to run the same experiment with two, three, or maybe even more different modalities (fMRI, PET with different tracers, etc).
This is incorrect, TUM has a PET scanner (site in German): https://nuklearmedizin.mri.tum.de/de/Patienten-Zuweiser/Pet-... Can't comment regarding the other observations.
And how it was almost impossible to reproduce many published and well cited result. It was both exciting and jarring to talk with the neuroscientist, because they ofc knew about this and knew how to read the papers but the one doing more funding/business side ofc didn't really spend much time putting emphasis on that.
One of the team presented a accepted paper that basically used Deep Learning (Attention) to predict images that a person was thinking of, from the fMRI signals. When I asked "but DL is proven to be able to find pattern even in random noise, so how can you be sure this is not just overfitting to artefact?" and there wasn't really any answer to that (or rather the publication didn't take that in to account, although that can be experimentally determined). Still, a month later I saw tech explore or some tech news writing an article about it, something like "AI can now read your brain" and the 1984 implications yada yada.
So this is indeed something probably most practitioners, masters and PhD, realize relatively early.
So now that someone says "you know mindfulness is proven to change your brainwaves?" I always add my story "yes, but the study was done with EEG, so I don't trust the scientific backing of it" (but anecdotally, it helps me)
So here you say quite a mouthful. If you train it on a pattern it'll see that pattern everywhere - think about the early "Deep Dream" trippy-dogs-pictures nonsense that was pervasive about eight or nine years ago.
I repaired a couple of cameras for someone who was working with a large university hospital about 15 years ago, where they were using admittedly 2010s-era "Deep Learning" to analyse biopsy scans for signs of cancer. It worked brilliantly, at least with the training materials, incredible hit rate, not too terrible false positive rate (no biggie, you're just trying to decide if you want to investigate further), really low false negative rate (if there was cancer it would spot it, for sure, and you don't want to miss that).
But in real-world patient data it went completely mental. The sample data was real-world patient data, too, but on "uncontrolled" patients, it was detecting cancer all over the place. It also detected cancer in pictures of the Oncology department lino floor, it detected cancer in a picture of a guy's ID badge, it detected cancer in a closeup of my car tyre, and it detected cancer in a photo of a grey overcast sky.
Aw no. Now what?
Well, that's why I looked at the camera for them. They'd photographed the biopsies with one camera on site, from "real patients", but a lot of the "clear" biopsies were from other sites.
You're ahead of me now, aren't you?
The "Deep Learning" system had in fact trained itself on a speck of shit on the sensor of one of the cameras, the one used for most of the "has cancer" biopsies and most of the "real patient under test" biopsies. If that little blob of about a dozen slightly darker pixels was present, then it must be cancer because that's what the grown-ups told it. The actual picture content was largely irrelevant because the blob was consistent across all of them.
I'm not too keen on AI in healthcare, not as a definitive "go/no-go" test thing.
You test your DL decoder on held-out data. This is the common practice.
Also saw this irl with a particular NGS diagnostic. This model was initially 99% accurate, P.I. smelled BS, had the grad student crunch the numbers again, 96% accurate, published it, built a company around this product —-> boom, 2 years later it was retracted because the data was a lot of amplified noise, spurious hits, overfitting.
I don’t know jack compared to the average HN contributor, but even I can smell the BS from a mile away in some of these biomedical AI models. Peer review is broken for highly-interdisciplinary research like this.
I also work in the field, specifically with sleep slow-wave enhancement.
Blood flow as a proxy for brain activity I always felt was a weak measure, as the brain activity involved across all manner of operating our biological systems, so is the increased blood flow measured in fMRI a response to cognition, or autonomic activity? What does that oxydation mean.
EEG is similarly flawed when we try to equate "brainwaves" to emotions and consciousness. I think we're almost better off measuring HRV, a much simpler measure, and more reliable.
I'm fascinated that so many people who discuss brainwaves think of them as actual "waves", when it is just how we plot electrical activity which creates a visual wave like pattern.
However, and this is specifically related to our work in sleep, we can detect slow-waves (I dislike that term, it's the synchronous firing of neurons) and we are able to stimulate this restorative brain function through sensory perception during sleep, and even create slow-waves in a lab using TMS.
Research linked on our website [1]
I agree the industry needs to stop projecting what we hope we're seeing with what is actually being measured, and we don't understand enough about how the brain works, but I think completely throwing away any brain related measures we have is going too far.
1 - https://affectablesleep.com/how-it-works#research
Wondering how they created that baseline. Was it with fMRI data (which has deviance from actual data, as pointed out)? Or was it through other means?
For task fMRI, the test-retest reliability is so poor it should probably be considered useless or bordering on pseudoscience, except for in some very limited cases like activation of the visual and/or auditory and/or motor cortex with certain kinds of clear stimuli. For resting-state fMRI (rs-fMRI), the reliabilities are a bit better, but also still generally extremely poor [1-3].
There are also two IMO major and devastating theoretical concerns re fMRI that IMO make the whole thing border on nonsense. One is the assumed relation between the BOLD signal and "activation", and two is the extremely horrible temporal resolution of fMRI.
It is typically assumed that the BOLD response (increased oxygen uptake) (1) corresponds to greater metabolic activity, and (2) increased metabolic activity corresponds to "activation" of those tissues. This trades dubiously on the meaning of "activation", often assuming "activation = excitatory", when we know in fact much metabolic activity is inhibitory. fMRI cannot distinguish between these things.
There are other deeper issues, in that it is not even clear to what extent the BOLD signal is from neurons at all (could be glia), and it is possible the BOLD signal must be interpreted differently in different brain regions, and that the usual analyses looking for a "spike" in BOLD activity are basically nonsense, since BOLD activity isn't even related to this at all, but rather the local field potential, instead. All this is reviewed in [4].
Re: temporal resolution, essentially, if you pay attention to what is going on in your mind, you know that a LOT of thought can happen in just 0.5 seconds (think of when you have a flash of insight that unifies a bunch of ideas). Or think of how quickly processing must be happening in order for us to process a movie or animation sequence where there are up to e.g. 10 cuts / shots within a single second. There is also just biological evidence that neurons take only milliseconds to spike, and that a sequence of spikes (well under 100ms) can convey meaningful information.
However, the lowest temporal resolutions (repetition times) in fMRI are only around 0.7 seconds. IMO this means that the ONLY way to analyze fMRI that makes sense is to see it as an emergent phenomenon that may be correlated with certain kinds of long-term activity reflecting cyclical BOLD patterns / low-frequency patterns of the BOLD response. I.e. rs-fMRI is the only fMRI that has ever made much sense a priori. The solution to this is maybe to combine EEG (extremely high temporal resolution, clear use in monitoring realtime brain changes like meditative states and in biofeedback training) with fMRI, as in e.g. [5]. But, it may still well be just the case fMRI remains mostly useless.
[1] Elliott, M. L., Knodt, A. R., Ireland, D., Morris, M. L., Poulton, R., Ramrakha, S., Sison, M. L., Moffitt, T. E., Caspi, A., & Hariri, A. R. (2020). What Is the Test-Retest Reliability of Common Task-Functional MRI Measures? New Empirical Evidence and a Meta-Analysis. Psychological Science, 31(7), 792–806. https://doi.org/10.1177/0956797620916786
[2] Herting, M. M., Gautam, P., Chen, Z., Mezher, A., & Vetter, N. C. (2018). Test-retest reliability of longitudinal task-based fMRI: Implications for developmental studies. Developmental Cognitive Neuroscience, 33, 17–26. https://doi.org/10.1016/j.dcn.2017.07.001
[3] Termenon, M., Jaillard, A., Delon-Martin, C., & Achard, S. (2016). Reliability of graph analysis of resting state fMRI using test-retest dataset from the Human Connectome Project. NeuroImage, 142, 172–187. https://doi.org/...
If they were to measure a person who performs mental arithmetic on a daily basis, I'd expect his brain activity and oxygen consumption to be lower than those of a person who never does it. How much difference would that make?
The question then is, do you expect a person who is really good at mental arithmetic to have less neural firing on arithmetic tasks (e.g., what is 147 x 38) than the average joe. I would hypothesize yes overall to solve each question; however, I'd also hypothesize the momentary max intensity of the expert to peak higher. Think of a bodybuilder vs. a SWE bench-pressing 100 lbs for 50 reps. The bodybuilder has way more muscle to devote to a single rep, and will likely finish the set in 20 seconds, while the SWE is going to take like 30 minutes ;)
fMRI is a cool, expensive tech, like so many others in genetics and other diagnostics. These technologies create good jobs ("doing well by doing good").
But as other comments point out, and practitioners know, their usefulness for patients is more dubious.
https://source.washu.edu/2025/12/psychedelics-disrupt-normal...
I think your expertise would be very welcome, but this comment is entirely unhelpful as-is. Saying there are bad comments in this thread and also that there is good literature out there without providing any specifics at all is just noise.
You don't have to respond to every comment you see to contribute to the discussion. At minimum, could you provide a hint for some literature you suggest reading?
Instead of a nothingburger, you could have used your academic prowess to break down the top 1/2 misconceptions with expertise.
You might not have time to respond to all the comments but a couple of clarifications could have helped anyone else who doesn't comment without experience.
Just saying that next time you can be the change you want to see in HN instead of wasting text telling us how ignorant we are.
To me this is like shitting on cars in 1925 because they kill people every now and then. Cars didn't go away, and nor will fMRI, until someone finds a better way to measure living people's brains.
TUM's press is being sloppy, from conflating fMRI with MRI to presuming this is revolutionary, and ignoring earlier empirical work against this narrative (Windkessel's, Logothetis beta/gamma coupling, etc.)
fMRI has always had folks highlighting how shaky the science is. It's not the strongest of experimental techniques.
We sped up fMRI analysis using distributed computing (MapReduce) and GPUs back in 2014.
Funny how nothing has changes.
What's still amazing is fMRI can provide more visual context of what's happening in the brain, in what region, and activities that can help that improve.
There are other complementary technologies like QEEG and SPECT that can also shed a light as well.
It does seem the case that fMRI cann be more of a snapshot photo, and technologies like SPECT can provide more of a regional time lapse of activity.
[just kidding]
1. http://prefrontal.org/blog/2009/01/voodoo-correlations-in-so...
2. https://journals.sagepub.com/doi/10.1111/j.1745-6924.2009.01...
3. https://www.mdpi.com/2076-0760/12/1/15
I know the actual diagnosis is several times more layered than this attempt at an explanation, but I always felt that trying to explain the brain by peering at it from outwards is like trying to debug code by looking at a motherboard through a bad microscope.