iPhone Targeted Content
iPad Targeted Content
Android Targeted Content
Blackberry Targeted Content
Desktop and all none targeted content
iPhone Targeted Content
THE CLINICAL CLARITY BLOG
Knees, Spines and the Bayesian Brain
Part 2
December 4, 2021 by DR MATTHEW D. LONG
iPad Targeted Content
Android Targeted Content
THE CLINICAL CLARITY BLOG
Knees, Spines and the Bayesian Brain
Part 2
December 4, 2021 by DR MATTHEW D. LONG
Blackberry Targeted Content
Desktop and all none targeted content
THE CLINICAL CLARITY BLOG
December 4, 2021 by DR MATTHEW D. LONG
In Part I of this article (here) we introduced the idea of prediction. In this instalment we'll investigate what prediction means within the context of musculoskeletal pain.
Our brain uses a vast storehouse of memories (data) to anticipate everything that we will see, smell, hear and feel. This is done subconsciously and constantly, allowing us to navigate the world safely and minimise potentially harmful surprises. Since birth our brains have steadily accumulated data about the intricacies of our body. But interoception (your brain's representation of sensations from your own body) is both noisy and imprecise. As such, the brain is constantly trying to predict what things mean and what's going to happen next.
According to Dr Lisa Feldman Barrett of Northeastern University (1), "The brain must continually construct concepts that guide the body by integrating scraps of sensory input with memories of similar experiences from the past. Creating this internal model of your body in the world allows the brain to infer the causes of the sense data that it receives through the retina and other sensory organs." In other words, it is the brain's inferences (hypotheses) about the origin of sensations, and not the sensory input itself, that determines our experience. Furthermore, the use of stored data becomes more important as the fidelity of the input signal decreases.
The authors Hechler et al (2) explain it thus,
Our brain uses a vast storehouse of memories (data) to anticipate everything that we will see, smell, hear and feel. This is done subconsciously and constantly, allowing us to navigate the world safely and minimise potentially harmful surprises. Since birth our brains have steadily accumulated data about the intricacies of our body. But interoception (your brain's representation of sensations from your own body) is both noisy and imprecise. As such, the brain is constantly trying to predict what things mean and what's going to happen next.
According to Dr Lisa Feldman Barrett of Northeastern University (1), "The brain must continually construct concepts that guide the body by integrating scraps of sensory input with memories of similar experiences from the past. Creating this internal model of your body in the world allows the brain to infer the causes of the sense data that it receives through the retina and other sensory organs." In other words, it is the brain's inferences (hypotheses) about the origin of sensations, and not the sensory input itself, that determines our experience. Furthermore, the use of stored data becomes more important as the fidelity of the input signal decreases.
The authors Hechler et al (2) explain it thus,
"While intuition suggests that sensations cause perception, recent evidence suggests that the brain predicts sensory input, so as to make inferences about the causes of the sensations. What we perceive therefore depends heavily upon the predictions of the brain, which reflect what the system already knows about the world and about the body. These predictions not only precede sensations, they determine sensation. Brains are thus conceived as prediction machines that function according to the Bayesian interpretation of probability that balance prior expectations against new sensory evidence."
The word 'Bayesian' above refers to a popular concept that the brain acts as a 'Bayesian probability machine' (3). According to this idea, the brain constantly makes predictions and then adjusts beliefs based on information provided by our senses. It then uses these refined inputs to construct a model of our world - both our external and internal worlds. The Bayesian brain is often described as an 'inference engine' that is always attempting to minimise prediction error. To demonstrate such prediction take a look at the video below, which illustrates the 'McGurk effect'.
Prediction occurs across the entire domain of bodily functions, not just our external senses. Our brain contains large datasets, or spatial maps, about the movement capabilities of the body (the so-called 'cortical body matrix'). Every time we move a body part a prediction is made, based upon our stored knowledge. We then use this prediction to keep us safe.
But how does this work in practice? According to Meier et al (4), the primary motor cortex (M1) does not function simply to issue motor commands. Rather, it also "models the proprioceptive consequences of motorneuron activity", thereby creating a virtual model of the intended movement before we execute the action. The final output of the alpha motorneurons is then determined after a comparison occurs within the spinal cord between the "descending proprioceptive predictions from M1 and the proprioceptive input from muscle spindle afferents." This allows us "to produce the desired (predicted) movement trajectory."
To clarify this idea, the brain wishes to minimise surprise, so it sends a prediction down to the spinal cord about the proprioceptive consequences that it expects to receive as movement begins. If there is an error in the prediction then the alpha motorneuron output is altered until the prediction error disappears. Not only that, but the proprioceptive feedback from muscle spindles is passed back to the sensory cortex (S1) to help the brain update its internal model. Projections from S1 to M1 allow the motor cortex to refine its predictions and improve performance of the system.
So what has this got to do with pain?
It appears that those with persistent pain often harbour changes within their somatosensory cortex. A paper by Roussel and colleagues (5) states that "altered cortical representation of somatic input may falsely signal an incongruence between motor intention and movement". In other words, a surprise. They also wrote,
But how does this work in practice? According to Meier et al (4), the primary motor cortex (M1) does not function simply to issue motor commands. Rather, it also "models the proprioceptive consequences of motorneuron activity", thereby creating a virtual model of the intended movement before we execute the action. The final output of the alpha motorneurons is then determined after a comparison occurs within the spinal cord between the "descending proprioceptive predictions from M1 and the proprioceptive input from muscle spindle afferents." This allows us "to produce the desired (predicted) movement trajectory."
To clarify this idea, the brain wishes to minimise surprise, so it sends a prediction down to the spinal cord about the proprioceptive consequences that it expects to receive as movement begins. If there is an error in the prediction then the alpha motorneuron output is altered until the prediction error disappears. Not only that, but the proprioceptive feedback from muscle spindles is passed back to the sensory cortex (S1) to help the brain update its internal model. Projections from S1 to M1 allow the motor cortex to refine its predictions and improve performance of the system.
So what has this got to do with pain?
It appears that those with persistent pain often harbour changes within their somatosensory cortex. A paper by Roussel and colleagues (5) states that "altered cortical representation of somatic input may falsely signal an incongruence between motor intention and movement". In other words, a surprise. They also wrote,
"Presenting incongruent information - a mismatch between intention, proprioception, and visual feedback or sensorimotor conflict - to healthy controls not only lead to an increased neuronal activity in right dorsolateral prefrontal cortex, but also induced pain and sensory disturbances in healthy controls and increased baseline symptoms in those with chronic pain. It has hence been proposed that a prolonged sensorimotor conflict may provoke long-term symptoms in healthy controls and that pain generated by this conflict may be considered as a warning signal to alert the individual to abnormalities within information processing."
Such notions about pain, and its origin within the higher centres, are often hard to appreciate. This is especially true for those who still think of pain as a "sensation" that arises from "damaged body tissues". But as we've seen, tissue injury is not necessary for someone to find themselves mired within the pain experience. Pain occurs when the body cannot reconcile prediction and feedback. Ongaro and Kaptchuk (6) summarise this nicely,
"We do not necessarily feel pain - this framework suggests - because we 'sense' it directly from the peripheral body. To put it emphatically, we feel pain because we predict that we are in pain, based on an integration of sensory inputs, prior experience, and contextual cues."
If the alteration to predicted inputs from our tissues is too great, and error increases, the brain must come up with a new hypothesis to explain the incoming data. Remember, our sense of discomfort is not determined simply by the sensory inputs themselves, but by hypotheses or inferences that the brain generates about the origin of the input. As time goes by some patients will move from an acute pain experience to chronic. As this occurs stiffness often ensues, as the brain's strategy shifts to one of reducing movement variability. The brain uses stiffness to try and minimise its prediction error, however, "matching the descending predictions from M1 to the reduced or disrupted proprioceptive input, probably provoke(s) neuroplastic adaptions in the long-term" (6). Furthermore, such rigid motor patterns are associated with greater spinal loading and degenerative change.
"The central principle underpinning this account is that physical symptoms, as felt and expressed by patients, are not a direct record of bodily activity, but an inference based on implicit predictions about interoceptive information, derived from prior knowledge." (7)
Let's now think about how this might unfold in our daily practice. Imagine a patient who has undergone an apparently successful total knee replacement (TKR), yet many months later they still experience pain. In desperation they consult with their original surgeon, who orders further imaging to ensure that the "prosthesis is still in place". As expected, there is nothing untoward found, and the individual is advised to "be patient" and continue with their rehabilitation. But the exercises are painful, and the sufferer worries that they are doing more harm than good. So they cease all exercise and try to get by, hoping that their pain will eventually disappear. What is going on here? Why might a patient continue to experience pain, in spite of well performed surgery?
To answer these questions we should recall our prior discussion of the the role of prediction in pain.
Prior to a total knee replacement the individual has spent their entire life amassing data about the performance characteristics of their knee. This allows them to accurately predict the sensory feedback that will occur whenever their knee is moved or subjected to strain forces. Over years they become 'expert' in the sensory signature of their own knee joint, and can quickly determine whether a new input is unexpected and possibly injurious.
Now imagine the situation when a patient awakens from their TKR surgery. The brain is forced to confront the loss of a significant portion of their sensory data (due to joint tissue removal), and their prediction accuracy is compromised. As discussed above, this will typically prompt a protective response - usually pain and stiffness - as their brain searches for clarity about their knee joint. Interestingly, the brain quickly learns that it can compensate for the loss of joint proprioception by recruiting or enhancing the mechanoreceptors embedded in the surrounding tissues (this even includes the skin). Slowly, over time, the brain can use this data to rebuild a sensory map of the knee, and update its prediction model for how the new knee moves. At least, this is what is supposed to happen. In some cases the brain does not seek to update its 'model' of the knee, and seemingly awaits the return of the old, familiar knee joint that it has grown up with. But this previous incarnation of the knee is never coming back, and it is up to the brain to adjust its understanding of the new knee, and alter its prediction model accordingly. Otherwise, the pain will last indefinitely.
So how do we help the brain to update its prediction model?
To answer these questions we should recall our prior discussion of the the role of prediction in pain.
Prior to a total knee replacement the individual has spent their entire life amassing data about the performance characteristics of their knee. This allows them to accurately predict the sensory feedback that will occur whenever their knee is moved or subjected to strain forces. Over years they become 'expert' in the sensory signature of their own knee joint, and can quickly determine whether a new input is unexpected and possibly injurious.
Now imagine the situation when a patient awakens from their TKR surgery. The brain is forced to confront the loss of a significant portion of their sensory data (due to joint tissue removal), and their prediction accuracy is compromised. As discussed above, this will typically prompt a protective response - usually pain and stiffness - as their brain searches for clarity about their knee joint. Interestingly, the brain quickly learns that it can compensate for the loss of joint proprioception by recruiting or enhancing the mechanoreceptors embedded in the surrounding tissues (this even includes the skin). Slowly, over time, the brain can use this data to rebuild a sensory map of the knee, and update its prediction model for how the new knee moves. At least, this is what is supposed to happen. In some cases the brain does not seek to update its 'model' of the knee, and seemingly awaits the return of the old, familiar knee joint that it has grown up with. But this previous incarnation of the knee is never coming back, and it is up to the brain to adjust its understanding of the new knee, and alter its prediction model accordingly. Otherwise, the pain will last indefinitely.
So how do we help the brain to update its prediction model?
We'll discuss this in the final instalment...
Dr Matthew D. Long
BSc (Syd), M.Chiro (Macq), DIANM
Dr Matthew D. Long
BSc (Syd), M.Chiro (Macq), DIANM
References:
1. Armstrong, Kim. Interoception: How We Understand Our Body’s Inner Sensations. Observer. Oct 2019. https://www.psychologicalscience.org/observer/interoception-how-we-understand-our-bodys-inner-sensations
2. Hechler, T., Endres, D., & Thorwart, A. (2016). Why Harmless Sensations Might Hurt in Individuals with Chronic Pain: About Heightened Prediction and Perception of Pain in the Mind. Frontiers in Psychology, 7, 1638. http://doi.org/10.3389/fpsyg.2016.01638
3. Raviv, Shaun. The Genius Neuroscientist Who Might Hold the Key to True AI. Wired Magazine. 13/11/18
4. Meier, M. L., Vrana, A., & Schweinhardt, P. (2019). Low Back Pain: The Potential Contribution of Supraspinal Motor Control and Proprioception. The Neuroscientist, 25(6), 583–596. http://doi.org/10.1177/1073858418809074
5. Roussel, N. A., Nijs, J., Meeus, M., Mylius, V., Fayt, C., & Oostendorp, R. (2013). Central Sensitization and Altered Central Pain Processing in Chronic Low Back Pain: Fact or Myth? The Clinical Journal of Pain, 29(7), 625–638. http://doi.org/10.1097/AJP.0b013e31826f9a71
6. Ongaro, G., & Kaptchuk, T. J. (2019). Symptom perception, placebo effects, and the Bayesian brain. Pain, 160(1), 1–4. http://doi.org/10.1097/j.pain.0000000000001367
7. Bergh, O. V. den, Witthöft, M., Petersen, S. & Brown, R. J. Symptoms and the body: Taking the inferential leap. Neurosci Biobehav Rev 74, 185–203 (2017).
1. Armstrong, Kim. Interoception: How We Understand Our Body’s Inner Sensations. Observer. Oct 2019. https://www.psychologicalscience.org/observer/interoception-how-we-understand-our-bodys-inner-sensations
2. Hechler, T., Endres, D., & Thorwart, A. (2016). Why Harmless Sensations Might Hurt in Individuals with Chronic Pain: About Heightened Prediction and Perception of Pain in the Mind. Frontiers in Psychology, 7, 1638. http://doi.org/10.3389/fpsyg.2016.01638
3. Raviv, Shaun. The Genius Neuroscientist Who Might Hold the Key to True AI. Wired Magazine. 13/11/18
4. Meier, M. L., Vrana, A., & Schweinhardt, P. (2019). Low Back Pain: The Potential Contribution of Supraspinal Motor Control and Proprioception. The Neuroscientist, 25(6), 583–596. http://doi.org/10.1177/1073858418809074
5. Roussel, N. A., Nijs, J., Meeus, M., Mylius, V., Fayt, C., & Oostendorp, R. (2013). Central Sensitization and Altered Central Pain Processing in Chronic Low Back Pain: Fact or Myth? The Clinical Journal of Pain, 29(7), 625–638. http://doi.org/10.1097/AJP.0b013e31826f9a71
6. Ongaro, G., & Kaptchuk, T. J. (2019). Symptom perception, placebo effects, and the Bayesian brain. Pain, 160(1), 1–4. http://doi.org/10.1097/j.pain.0000000000001367
7. Bergh, O. V. den, Witthöft, M., Petersen, S. & Brown, R. J. Symptoms and the body: Taking the inferential leap. Neurosci Biobehav Rev 74, 185–203 (2017).