A structural causal model for robot-assisted upper-limb neurorehabilitation

Area of Science:

• Neuroscience
• Robotics
• Causal Inference

Background:

• Robot-assisted neurorehabilitation for stroke shows limited functional gains and high patient variability.
• Current approaches often rely on correlational findings rather than interventional reasoning.
• Precision neurorehabilitation necessitates mechanistically driven, tailored therapy models.

Purpose of the Study:

• To present a structural causal model (SCM) for upper-limb robot-assisted neurorehabilitation.
• To encode known and hypothesized causal relationships in robot-assisted therapy using a directed acyclic graph (DAG).
• To demonstrate how the SCM can explain observed phenomena and generate testable interventional predictions.

Main Methods:

• Development of a directed acyclic graph (DAG) representing key constructs in robot-assisted neurorehabilitation.
• Encoding of causal influences between nodes based on existing literature.
• Analysis of the DAG's ability to account for observed phenomena and predict interventional effects.

Main Results:

• The proposed causal graph successfully models several observed phenomena in robot-assisted therapy.
• The model yields testable predictions regarding the effects of specific interventions.
• A conceptual example illustrates how the SCM can guide decisions on therapy parameters for individual patients.

Conclusions:

• The developed causal graph provides a framework for interventional reasoning in robot-assisted neurorehabilitation.
• Empirical investigation is required to validate and refine the causal structure.
• This SCM has the potential to advance mechanistic understanding and improve patient outcomes in robot-assisted therapy.