Functional neurological disorder: new subtypes and shared mechanisms, 2022, Hallett, Dworetzky, Stone et al

Slightly OT but stumbled across this alternative take on "PEM" while looking to update my understanding of predictive coding/processing. My understanding remains vague but I can see how a superficial reading of selected parts of the model could be misused to lend support to all manner of mind-body mischief.

Anyway, "PEM":

https://predictive-mind.net/papers/vanilla-pp-for-philosophers-a-primer-on-predictive-processing

A slightly shorter and less dense - if overenthusiastic - explanation of predictive coding/processing if anyone's interested:
https://www.mindcoolness.com/blog/bayesian-brain-predictive-processing/

 

I suggest this is because they are collaborating with younger scientists who are conducting those neuroimaging studies and may have slightly different opinions.

@dave30th

More to the point, they're merely speculating that there is a predictive system of sense inputs other than proprioception, without any empirical evidence, despite a century of research trying to look for such evidence.

The brain doesn't need to predict fatigue or pain because a small amount of latency in those signals doesn't matter at all. Instead of a predictive coding system, we have a reflex system.

There is a predictive coding system for proprioception, because our nerves are too slow to provide feedback in real time to provide precise motor control. (this research dates back to Helmholtz in the mid 19th century https://en.wikipedia.org/wiki/Predictive_coding)

Their system is backwards because instead of relying more on sensory input to correct the model when it is incorrect, they are instead suggesting that the system decides to ignore sensory input and just mark one's own homework as 100% correct. This would necessarily lead to a loss in precise motor control (and hence does not apply to ME/CFS patients). They are also assuming that the brain doesn't have multiple sensory pathways to correct for when muscle (force/velocity) afferent signals may be incorrect - starting with the signals from the surrounding muscles, but also visual and touch feedback. If their model is correct, then all of these additional feedback mechanisms also have to be in error or are ignored by the brain.

 

That is a reasonable point but predictive coding applies much more widely than proprioception. It is relevant to colour and shade vision and to sounds, including speech perception. If the brain predicts that a square on a picture of a chessboard is in the shade of a vase then it will interpret it as white even though the shade of the image is the same as for black squares out of shadow.

 

I wouldn't strictly describe that as "predictive coding" though, as the reasons are more complex.
Some of our perceptual processing related to hearing and sight are deliberately lossy (in terms of information processing) for example, in ways that aren't strictly related to accurate prediction of future sensory input. To put it another way, the brain filters the inputs to focus on what is deemed to be meaningful such as separating noise from speech, edge finding to provide spatial awareness and motion tracking which is needed for motor control of the eye, etc. The function is not to accurately predict the totality of the sensory input.

This contrasts with the "perceptual coding" errors proposed by Stone et al. who flip the whole concept backwards and claim the higher level systems try to predict the noise rather than the meaningful signals and then lose the plot.

 

It is what the predictive coding people call predictive coding.


I spend most of my time when not thinking about ME thinking about perception and writing articles on it. The predictive coding people are the main neurobiological theorists involved. They consider the brain to be a prediction machine with everything in perception derived from a 'Bayesian model', whatever that is supposed to be. Predictive coding is supposed to explain why we are conscious at all. It is blather, but it is the current fashion.

There is a vast literature on illusions that fool the 'prediction machine' including colour and shape illusions, rubber hands and misinterpretations of lip movements and so on. There is masses of evidence for 'forward models' and 'efferent copies' and so on although as far as I am aware nobody can say how it actually works in individual neurons. My work is on how perception might operate at the individual neuron level. I do not doubt that comparisons and subtractions are widespread in all modalities but I don't think the neurobiologists realise quite what that needs to involve. Christof Koch and Idan Segev got somewhere near to it in giant locust visual neurons about twenty years ago but nothing much new seems to have materialised.

So I don't think people like Mark Edwards are thinking in terms of proprioception particularly when they talk of predictive coding.

 

Why would they not think about the most strongly empirically demonstrated biological predictive coding systems? (I now realise this is a rhetorical question)
That is where the language of 'forward models' and 'efferent copies' comes from (I'm a fan of Helmholz's work).

The idea is that the brain (supplementary motor areas) has an effective model of the muscle efferents and specific afferents, such that when a motor signal is generated, an identical the 'efferent copy' is passed through, it provides a precise series of outputs that are then compared with the actual afferent outputs, which in turn revises the model. (of course there are other inputs too, vision, vestibular system etc)

Knowing little about the field, and thinking about bottoms-up approaches, I'd look at models of force loaded spring-mass systems based on neuronal modelling and then construct a feedback-control system with separate hierarchies (and deliberately apply latency). But then again, this could lead to a system being created that despite being built out of neurons and delivers the theoretical capability, has nothing in common with actual biological brains.

I don't know if anyone has tried to develop the canonical models as actual neuronal systems (https://www.sciencedirect.com/science/article/pii/S0896627312009592).

There is this also:
"Single cortical neurons as deep artificial neural networks"
https://www.sciencedirect.com/science/article/abs/pii/S0896627321005018?dgcid=author

Perhaps start a different thread if this is something worth talking about?

 

First author video “Physiology of free will” haven’t watched

https://www.youtube.com/watch?v=6iasgavh8uo

 

Anyone have access to the full article? Wondering who funded this.

 

The majority of the talk had little to do with functional motor disorders. The video mostly focused on the fact that our sense of attention or 'volition' during a motor task occurs after the task actually occurs.

Several signals are observed in the brain during a motor task, firstly a readiness potential that basically lets the brain know a motor task may follow, this is typically around 1 second before but can be up to 8 seconds during some tasks. Then there are several senses of a generation of a task, with the sense of will occurring after the time at which the motor task occurred, yet the individual percieves it as being around 200-300ms before the task. They also devised an experiment that attempted to predict a task based on the readiness potential, then provide feedback to the participant to prevent the task. The 'point of no return' was around 200ms before the task occurred, however many participants quickly aborted the task partway through. He also mentioned that Transcranial Magnetic Stimulation of the supplementary motor areas 200 ms after a task has started can disrupt an individuals sense of when the task occurred. Additional note from me: TMS of the SMA also disrupts the sense of effort during motor tasks.

The question of functional motor disorders was covered in the first question, but the answer was about why individuals don't perceive there to be a sense of volition associated with the motor movement. Hallett suggests the problem is in the right temporoparetial junction that has been shown to be integrate our conscious attention and sense of volition. But he didn't answer as to why this system may be going wrong, nor why the movements are generated in the first place.

 

Comment: Functional neurological disorder and other unexplained syndromes

"Functional neurological disorder is a syndrome of medically unexplained neurological symptoms. In The Lancet Neurology, Mark Hallett and colleagues review some of the potential explanations for functional neurological disorder and the evidence that supports these explanations. 1

The paper by Hallett and colleagues, however, is more than a Review: it is also a territorial claim, seeking to expand the boundaries of what should be considered functional neurological disorder. The details of this claim are unlikely to be controversial to any clinician working in the field: the presentations Hallett and colleagues describe are not new, even if they do not fall within the current classifications of the disorder. But the claim is nonetheless remarkable, as even a decade ago it would have been thought to be sheer folly. A good argument could then have been made that functional neurological disorder (or conversion disorder, as it was more formally known) was the most stigmatised of all disorders, even compared with other unexplained syndromes. What would have been the point of expanding the scope of a diagnosis that patients went to such lengths to avoid? 2

The expansive mood in the Review by Hallett and colleagues therefore reflects a striking transformation in the status of functional neurological disorder. Functional neurological disorder has become a diagnosis that a neurologist might be comfortable to give, and that a patient might be glad to receive."

Paywall, https://www.thelancet.com/journals/laneur/article/PIIS1474-4422(22)00095-3/fulltext

 

The author of this fawning dross is Richard Kanaan. Not surprisingly, he and Wessely are old cronies.

I believe the technical term for this is 'circle jerk'.

 

Has anyone read the actual paper? I cannot access through sci-hub. Thanks.

 

When I was first diagnosed with FND I was directed to Stone’s website.

All the consultant herself said to me as an explanation for the diagnosis was some very brief waffle about software, as per Stone. Then she told me to check the website out.

I was instantly suspicious. Then I checked out the website and got much more suspicious.

 

Ah so patients are all Buddhist monks ...
( power of mind over physical )

 

About this bit quoted by @Art Vandelay above from Jon Stone's website about Vogt's recovery stories

Have they? I was surrounded with them. Bombarded by tales of people doing the same things as me and just having different outcomes.

One of the most dispiriting experiences I had was reading a book about 50 supposed recoveries. It was so clear there were no patterns, that these people had spontaneously improved. The only things I took away were, ensure I don't have something else like untreated hypothyroidism, and consider diet. Neither helped. I get that some people find recovery stories inspiring; for me, they have the opposite effect. Look! Another person has improved! And it's not you! And they attribute it to things you've done or homeopathy! Super.

 

From a UK perspective ...

Be aware that diagnosis of hypothyroidism is usually restricted to just diagnosing Primary Hypothyroidism (PH). Other forms of hypothyroidism are rarely tested for and rarely treated because doctors are trained that such other forms are rare and thus belong in the "zebra" category which is often ignored. Also note that if a patient has PH they can still be missed for many years because doctors seem to rarely think of it - I don't know why.

Also testing and dose levels are usually determined by TSH alone these days because "TSH is the gold standard marker for thyroid disease" and setting doses for treatment. And - "If the TSH is in range then the thyroid hormone levels must be okay so we don't have to test them". This is nonsense and leaves many people struggling for years without treatment.

TSH is produced by the pituitary. But doctors think pituitary problems are vanishingly rare so can again be ignored. There are oodles of thyroid forums on the web if it interests you.

 

Just having a look at this older thread as I was reviewing previous comments on predictive coding.

That paper was the commentary article on the thread's main paper. Here's a few quotes —

The closing sentence referencing Courage for sound science wins John Maddox prize (2012, Nature)

 

I want to summarise for myself, which I hope may also be useful for other later readers of this thread. If I'm following correctly the proposals are —

The brain is constantly predicting the environment. It may do this for all sensory inputs but this is theoretical and we don't have evidence for this beyond established things like being too fast for eg visual nerve conduction velocities. Allowing the ability to predict a ball's motion and move a hand to catch it faster than the time taken to receive/process/send the totality of visual/audio/motor signals.

https://www.youtube.com/watch?v=DKMllY6jHp0

The prediction is checked and if the real input signals are within satisfactory error, the model simply remains as "truth". If the input signals are out-of-error-bounds, the model is discarded and updated. So that is a mechanism of regulation claimed for the normal situation.

But BPS says "your model's totes out of whack my dude", making the prediction wrong (because "false beliefs") and that's why you are expecting pain, fatigue, whatever.

But as Jo says the real inputs are claimed to be zero pain, zero fatigue (because they say there's nothing actually biologically wrong in the periphery to give a positive signal). So the above mechanism should be invoked, the model refreshed with the bottom line result of "akshully no pain, fatigue" (or even minus pain, fatigue).

But BPS are claiming that the outcome is still pain and fatigue, as initially predicted and the model is ignoring the actual inputs. So the update mechanism is not invoked. So now two things are wrong: initial model parameters (I hurt / will experience fatigue); and no error-correction.

So we're left with (as Jo says) a requirement for how this works normally, but also how it now fails / is inverted in illness. In which case "false beliefs" or indeed any psychological inputs are irrelevant, because the mechanism of creating and updating the model is broken. (Which if true sounds like there might be a structural problem...)

If the model-error-correction mechanism is indeed broken, as per the above requirement, you could fix it by simply telling yourself "no pain", "no fatigue" and the model would just stick with it. (This is LP/CBT/GET/ANS rewire etc). Lots of credulous people are very keen on the idea of brain retraining - why aren't they all fixed just by convincing themselves "no pain"?

It can't be that the brain retraining also repairs the model-error-correction bug, as again the problem would be fixed regardless of the person's prior ideas. Or you would now need yet another failure mode where

initial_model = "no pain" + zero sensory input => "pain".

In other words you might need a third explanation for why these mechanisms don't apply in failed cure.

I hope I've got the broad strokes right.

 

I think that analysis is a misunderstanding. Predictive coding has nothing to do with explaining how we do things 'too fast'. That is an argument about the role of conscious percepts, which occur too late to explain what the brain does in action. The nerve pathways are fast enough. It is just that the conscious aspect that we like to think is the cause of our 'free will' in practice cannot be. Our will works at a lower level of processing.