Cerebellar Ataxia – Research Application

[Masato Kobayashi]

One of the articles mentioned in the cerebellar dysfunction module (Therrien, A. S., Wolpert, D. M., & Bastian, A. J. (2016). Effective reinforcement learning following cerebellar damage requires a balance between exploration and motor noise. Brain : a journal of neurology, 139(Pt 1), 101–114. https://doi.org/10.1093/brain/awv329) discussed use of reinforcement learning (success vs failure/knowledge of results) for patients with cerebellar dysfunction. I wanted to open dialogue with everyone on how we incorporate this into clinical practice. In the study, they used a very detailed reaching task, but I would imagine this could be especially difficult to apply to more dynamic, multi-joint movements like sit to stands. Would it be as straight forward as not providing feedback on how they perform a sit to stand but more so on if their sit to stand achieves a certain desired outcome (i.e. presence or absence of loss of balance at the final extension phase with verbal confirmation of outcome?). Sorry if this is a long-winded post! Just wanted to pick everyone’s brain.

[Chrissy Durrough Lugge]

Hi Masato! I love this conversation!

Here’s a little more information from the authors. The first download under Supplementary Material at this link provides a full list of the exercises: https://www.neurology.org/doi/10.1212/WNL.0b013e3181c33adf#supplementary-materials

The whole body movements included arm and leg movements in standing and kneeling activities.

Error-based learning requires feedback about error data to allow the learner to make corrections on a trial by trial basis. If you recall our motor learning module, error-based learning requires knowledge of performance and often includes concurrent feedback. In the clinic, this might look like using biofeedback about weight shifting on a forceplate to control a figure on a computer monitor. It could also be throwing Velcro balls at a target where the learner can see how far their attempts land from the bullseye.

Reinforcement learning involves only knowledge of results (success or failure) and terminal feedback that forces the learner to explore options to facilitate more successful outcomes. In the clinic, this might look like throwing a ball at a target and the learner only knowing whether they hit the target (success) or not (failure).

Closed loop reinforcement learning means that task difficulty is controlled based on recent performance. In the ball-throwing example, this might mean moving the target to make it easier or harder to hit based on the patient’s recent attempts.

For sit-to-stands, I would imagine distilling the outcome down to one facet that determines success/failure (like balance during the final extension phase) would successfully implement reinforcement learning. The task can then be modified based on recent performance by altering seat height, the surface on which the feet are placed (stable vs. unstable), etc.

Interested to hear what everyone else thinks! 🙂

Chrissy

[Author]

Posts