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

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• General Movement Assessment (GMA) is a non-invasive method for early detection of brain dysfunction.
• Automated GMA methods can expand clinical applications but face challenges with data noise and pose estimation uncertainty.
• Current methods lack reliable uncertainty quantification, limiting clinical trust.

Purpose of the Study:

• To develop an automated General Movement Assessment (GMA) method that addresses uncertainty in pose-based analysis.
• To enhance the reliability and clinical applicability of automated GMA through robust uncertainty modeling.

Main Methods:

• Introduced UDF-GMA, a novel approach for pose-based automated GMA.
• Explicitly modeled epistemic uncertainty (model parameters) and aleatoric uncertainty (data noise).
• Employed Bayesian approximation for epistemic uncertainty estimation and direct modeling for aleatoric uncertainty.
• Fused disentangled uncertainties with embedded motion representation to improve class separation.

Main Results:

• The UDF-GMA method demonstrated effectiveness in predicting poor repertoire.
• Experiments on the Pmi-GMA benchmark dataset confirmed the approach's generalisability.
• The proposed method successfully disentangled and utilized both epistemic and aleatoric uncertainties.

Conclusions:

• UDF-GMA offers a reliable solution for uncertainty-aware automated General Movement Assessment.
• The method has the potential to significantly improve early detection of brain dysfunction.
• This work advances automated GMA by integrating robust uncertainty quantification for enhanced clinical utility.