A Shape-Based Functional Index for Objective Assessment of Pediatric Motor Function

Technical Summary: A Shape-Based Functional Index for Objective Assessment of Pediatric Motor Function

Problem Statement
Clinical assessments for neuromuscular disorders, specifically Duchenne Muscular Dystrophy (DMD) and Spinal Muscular Atrophy (SMA), currently rely heavily on subjective, observational measures (e.g., Brooke Upper Extremity Scale, CHOP-Intend) to monitor disease progression and treatment response. These methods are prone to clinical bias, lack precision in capturing subtle changes, and are limited to sporadic clinical visits. Furthermore, pediatric movement data presents unique analytical challenges due to confounding factors such as limb length variations in growing children, variability in movement speed, and differing developmental abilities. Existing digital medicine classifiers often rely on "black-box" features that lack interpretability for clinicians. The authors aim to develop an objective, data-driven method using wearable sensors to assess motor function in home settings, capable of handling temporal misalignment and providing interpretable kinematic patterns.

Methodology
The study proposes a pipeline combining wearable sensor data with Shape-based Principal Component Analysis (Shape PCA) and Partial Least Squares (PLS) regression.

1. Data Collection:

   • Participants: 31 participants (15 DMD, 7 SMA, 9 Healthy controls) recruited from the University of Virginia Pediatric Neuromuscular Clinic.
   • Sensors: MetaMotionR+ sensors (accelerometer and gyroscope at 200 Hz) worn on dominant and non-dominant hands.
   • Tasks: Activities of Daily Living (ADLs) including arm curls, door knob rotation, cup raising, door knocking, and paddle moving.
   • Clinical Measures: Brooke scores, ultrasound measures (Cross-Sectional Area, Average Echogenicity), and dynamometer measures (Normalized Elbow Torque).

2. Curve Registration and Phase-Amplitude Separation:

   • The authors utilize Square-Root Velocity Functions (SRVFs) to perform temporal alignment of movement trajectories. This approach separates the "phase" (timing/speed) from the "amplitude" (shape) of the motion.
   • Using an elastic approach and Dynamic Programming Algorithm (DPA), individual curves are aligned to a mean shape, removing the influence of movement speed variations. This allows for the comparison of biomechanical signals independent of the rate at which activities are performed.

3. Shape-Based Principal Component Analysis (Shape PCA):

   • After alignment, the authors perform PCA in the SRVF space (Shape Space) to identify principal modes of variation (Vertical Principal Components or VPCs).
   • This step isolates interpretable kinematic patterns, such as scaling (speed) and asymmetry, distinct from noise or timing differences.

4. Statistical Modeling:

   • Bootstrapping: To assess variability, 10,000 bootstrap replicates were generated to evaluate the stability of relationships between wearable features and clinical outcomes.
   • Partial Least Squares (PLS): PLS was employed to combine projections on the principal components (VPCs) and identify a covariation mode that correlates with clinical measures (muscle fat infiltration, Brooke scores, age-related degeneration).
   • Mixed Linear Models: Used to analyze cross-sectional trends regarding age and cohort interactions.

Key Results

• Kinematic Patterns: Shape PCA identified distinct modes of variation:
  • VPC1 (Speed): Primarily reflects changes in angular velocity (scaling). It explained 53.87% of variance in arm curl data.
  • VPC2 (Asymmetry): Captures asymmetry in motion (e.g., differences between upward and downward phases). It explained 25.5% of variance.
  • VPC3: Represented sensor or segmentation noise.
• Cohort Differences:
  • Both DMD and SMA cohorts showed higher Average Echogenicity (indicating fat infiltration) and lower Normalized Elbow Torque compared to healthy controls.
  • SMA Specificity: Patients with SMA exhibited significantly greater activation of the motion asymmetry pattern (VPC2 Curl) compared to DMD and healthy controls, suggesting SMA may impair subtle motion control differently than DMD.
  • Both patient groups showed reduced speed in knocking motions compared to healthy controls.
• Functional Index Construction:
  • The PLS analysis identified a primary canonical dimension combining speed and asymmetry features: $0.76 \times \text{speed}_{\text{curl}} - 0.59 \times \text{speed}_{\text{knock}} + 0.17 \times \text{asymmetry}_{\text{curl}} + 0.18 \times \text{asymmetry}_{\text{knock}}$.
  • This dimension achieved a canonical correlation of $r = 0.78$ (95% CI: [0.34, 0.94]) with muscle fat infiltration, Brooke scores, and age-related degenerative changes.
  • The proposed Shape PCA method outperformed Functional PCA (without alignment) and Non-negative Matrix Factorization (NMF) in terms of median canonical correlation and confidence interval stability.
• Longitudinal Trends: Mixed-effects modeling revealed a significant age-related decline in movement speed (positive slope coefficients) for both DMD ($\beta = 1.337$) and SMA ($\beta = 2.530$) cohorts, whereas healthy controls showed no significant temporal loss of function.

Significance and Claims
The paper claims to introduce a novel, interpretable motor function index derived from wearable sensor data that addresses the limitations of current subjective assessments. Key claims include:

• Objectivity and Interpretability: Unlike black-box classifiers, the Shape PCA approach yields transparent, intuitive results (e.g., specific asymmetry patterns) that clinicians can understand and trust.
• Robustness to Confounders: The phase-amplitude separation method effectively handles the complexities of pediatric data, such as limb length variations and inconsistent movement speeds, allowing for accurate trajectory comparison.
• Home-Based Monitoring Potential: The method is designed for deployment in home settings, enabling continuous, longitudinal tracking of treatment efficacy and disease progression without the burden of frequent clinical visits.
• Clinical Relevance: The derived index correlates strongly with established clinical metrics (muscle fat infiltration, strength, functional scores), suggesting it can serve as a sensitive biomarker for evaluating therapeutic interventions in rare neuromuscular diseases.

The authors conclude that this data-driven approach offers a practical pathway for improving patient monitoring and clinical decision-making, particularly for rare diseases where recruiting large cohorts for traditional trials is difficult.