UDF-GMA: Uncertainty Disentanglement and Fusion for General Movement Assessment

2025-07-07

AI Summary

General motion assessment (GMA) based on pose estimation suffers from high model uncertainty and unreliable predictions due to scarce high-quality annotated data and pervasive pose estimation noise. Method: This work introduces the first explicit decoupling and joint modeling of epistemic uncertainty (approximated via Bayesian inference) and aleatoric uncertainty (directly modeled) within the GMA framework, integrating both into motion embedding representations to enhance class discriminability. Contribution/Results: The proposed uncertainty fusion mechanism significantly improves robustness and discriminative capability—particularly for low-quality or impoverished motion patterns. Evaluated on the Pmi-GMA benchmark, our approach achieves substantial gains in prediction accuracy, demonstrating strong effectiveness, generalizability across diverse motion conditions, and practical potential for clinical deployment.

Technology Category

• Reasoning under Uncertainty: Uncertainty Representations
• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Motion & Tracking

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

General movement assessment (GMA) is a non-invasive tool for the early detection of brain dysfunction through the qualitative assessment of general movements, and the development of automated methods can broaden its application. However, mainstream pose-based automated GMA methods are prone to uncertainty due to limited high-quality data and noisy pose estimation, hindering clinical reliability without reliable uncertainty measures. In this work, we introduce UDF-GMA which explicitly models epistemic uncertainty in model parameters and aleatoric uncertainty from data noise for pose-based automated GMA. UDF-GMA effectively disentangles uncertainties by directly modelling aleatoric uncertainty and estimating epistemic uncertainty through Bayesian approximation. We further propose fusing these uncertainties with the embedded motion representation to enhance class separation. Extensive experiments on the Pmi-GMA benchmark dataset demonstrate the effectiveness and generalisability of the proposed approach in predicting poor repertoire.

Problem

• Automated GMA hindered by uncertainty from data and pose estimation
• Lack of reliable uncertainty measures reduces clinical reliability
• Need to disentangle and fuse uncertainties for better movement assessment

Innovation

• Models epistemic and aleatoric uncertainties explicitly
• Disentangles uncertainties via Bayesian approximation
• Fuses uncertainties with motion representation