Multiscale SPD manifold learning for rehabilitation exercise evaluation

Area of Science:

• Biomedical Engineering
• Computer Science
• Data Science

Background:

• Rehabilitation exercise assessment is vital for patient recovery, especially after mobility-affecting events.
• Traditional Euclidean methods struggle to capture complex motion variations and spatial relationships in rehabilitation data.
• Skeleton-based data from rehabilitation exercises requires advanced analytical techniques for accurate assessment.

Purpose of the Study:

• To propose a novel framework for rehabilitation exercise assessment using symmetric positive definite (SPD) manifolds.
• To leverage the geometric properties of SPD manifolds for preserving intrinsic human motion characteristics.
• To improve the accuracy and efficiency of classifying correct versus incorrect rehabilitation movements.

Main Methods:

• Representing skeleton-based rehabilitation data as points on a symmetric positive definite (SPD) manifold.
• Integrating unsupervised K-Nearest Neighbors (KNN) with Riemannian geometry for movement classification.
• Developing a Tangent Space Linear SPD Support Vector Machine (SVM) optimized via stochastic gradient descent (SGD).
• Designing a specialized neural network with multi-scale feature extraction for vectorized SPD data.

Main Results:

• The proposed SPD manifold approach significantly outperforms existing methods on benchmark datasets (Kimore, UI-PRMD, EHE).
• Achieved high cross-subject accuracies: 92.40% (UI-PRMD), 85.18% (Kimore), and 87.59% (EHE).
• Demonstrated faster training convergence and reduced computational overhead compared to state-of-the-art techniques.

Conclusions:

• Symmetric positive definite (SPD) manifolds offer a powerful tool for accurate and reliable rehabilitation exercise assessment.
• The proposed framework effectively captures intrinsic geometric structures and nonlinear variations in human motion.
• This novel approach has the potential to significantly enhance patient recovery monitoring and therapeutic guidance.