Enhancing Clinician Decision-Making via Uncertainty-Aware Multi-Expert Fusion for Stroke Rehabilitation

2026-06-23

AI Summary

Current stroke rehabilitation assessment methods, such as the Action Research Arm Test (ARAT), reduce complex motor behaviors to a single ordinal score, making it difficult to distinguish genuine recovery from compensatory strategies. Moreover, automated systems often suffer from label noise and lack of interpretability, limiting their clinical adoption. This work proposes xAARA, a system that leverages multi-view video inputs and integrates 692 calibrated multimodal expert models via a dynamic Bayesian network to enable three-tiered evaluation of tasks, movement phases, and movement quality. An entropy-gating mechanism dynamically weights model outputs and abstains from predictions under low confidence. By incorporating uncertainty quantification and multilevel clinical interpretability into rehabilitation assessment for the first time, xAARA achieves 94.2% task-level accuracy (κ=0.934) and 81.3% phase-level accuracy (κ=0.727) across 788 exercises from 105 patients, reduces prediction uncertainty by 96.1% compared to individual clinician ratings, and has been endorsed by four clinicians for clinical use.

Abstract

Tailoring stroke rehabilitation requires assessing how movements are organized, not merely if they succeed. Currently, this assessment is a rate-limiting bottleneck. Instruments like the Action Research Arm Test (ARAT) compress rich behavioral observations into single ordinal endpoints, discarding the movement-quality details that distinguish recovery from compensation. Automated alternatives typically chase accuracy on noisy, single-observer labels to output opaque scores - a technology-centric approach that rarely reaches clinical practice. To address this, we present xAARA: an engine designed to augment rather than replace clinical judgment. From multi-view video, xAARA returns ARAT assessments with calibrated uncertainty and explanations across task, movement-phase, and movement-quality levels. Treating clinical scoring as an ill-posed inference problem, xAARA composes 692 calibrated multimodal models via a Dynamic Bayesian Network with entropy-based gating. It qualifies results against clinical validity rules and defers low-confidence cases. In 105 stroke survivors (788 exercises), xAARA achieved 94.2% task accuracy (Cohen's kappa=0.934) and 81.3% movement-phase accuracy (kappa=0.727), reducing predictive uncertainty by 96.1% compared to single-clinician scoring. For subjective cases, it matched at least one rater 100% of the time and never returned out-of-range scores. Four independent clinicians validated the assessments and indicated willingness to adopt the system. We argue that principled uncertainty quantification and clinician-aligned explainability are the critical bridges moving automated assessment from technical demonstration to a deployable clinical tool.

Problem

• stroke rehabilitation
• clinical decision-making
• movement quality
• assessment bottleneck
• uncertainty quantification

Innovation

• Uncertainty Quantification
• Multi-Expert Fusion
• Explainable AI
• Dynamic Bayesian Network
• Clinical Decision Support