Using principal component analysis to reduce complex datasets produced by robotic technology in healthy participants

Area of Science:

• Robotics and Biomechanics
• Neuroscience
• Data Science

Background:

• The KINARM robot generates over 100 performance metrics per participant, covering proprioceptive, motor, visuospatial, and executive functions.
• This high dimensionality poses challenges for clinical interpretation of complex sensorimotor data.

Purpose of the Study:

• To reduce the multivariate data generated by the KINARM robot using Principal Component Analysis (PCA).
• To increase the interpretability of KINARM performance metrics while minimizing information loss.

Main Methods:

• Healthy, right-hand dominant participants (N=101-208) were assessed using a bilateral KINARM end-point robot across 6 behavioral tasks.
• Performance metrics (9-20 per task) were converted to Z-scores and subjected to PCA.
• Component selection was guided by scree plots, parallel analysis, and interpretability criteria.

Main Results:

• PCA substantially reduced KINARM data dimensionality (67-79% reduction per task) while retaining significant variance (70-82%).
• Most tasks yielded 3 principal components, with one task yielding 5 components.
• High loadings and minimal cross-loadings indicated strong component separation.

Conclusions:

• Principal Component Analysis (PCA) effectively reduces complex KINARM robot data into a smaller set of interpretable components.
• This data reduction technique enhances the clinical interpretability of sensorimotor performance metrics.
• Future applications of PCA may reveal specific patterns of sensorimotor impairment in patient populations.