Principal Components Analysis Using Data Collected From Healthy Individuals on Two Robotic Assessment Platforms...

Area of Science:

• Biomechanics
• Robotics
• Data Science

Background:

• Kinarm Standard Tests (KSTs) generate over 100 variables per individual, assessing upper limb sensory, motor, and cognitive functions.
• Previous studies utilized Principal Component Analysis (PCA) for data reduction with the Kinarm End-Point Lab (EP).
• This study investigates PCA on data from the Kinarm Exoskeleton Lab (EXO) and platform agreement.

Purpose of the Study:

• To perform PCA on Kinarm Exoskeleton Lab (EXO) data.
• To determine the agreement of PCA results between EXO and EP platforms in healthy participants.
• To explore further dimensionality reduction by applying PCA across multiple behavioral tasks.

Main Methods:

• Healthy participants (N=469 for EXO, N=170-200 for EP) performed four tasks quantifying arm sensory and motor function.
• PCA was applied per task, with component selection based on scree plots and parallel analysis.
• Agreement between platforms was assessed using distance correlation; inter-task PCA was also performed on EXO data.

Main Results:

• PCA on EXO data reduced parameters by 58-75%, explaining 76-87% of variance.
• Good-to-excellent agreement (0.75-0.99) was found between principal components (PCs) from EXO and EP platforms.
• Dimensionality reduction across tasks yielded 16 components from 76 parameters (79% reduction), explaining 73% of variance.

Conclusions:

• PCA effectively captures kinematic feature relationships in healthy individuals.
• The PCA approach is robust and platform-agnostic for Kinarm robotic assessments.
• Future research should explore PCA for characterizing neurological deficits in clinical populations.