Development and Verification of Postural Control Assessment Using Deep-Learning-Based Pose Estimators: Towards...

Area of Science:

• Biomechanics
• Rehabilitation Engineering
• Artificial Intelligence in Healthcare

Background:

• Postural control is a fundamental aspect of occupational performance evaluated by therapists.
• Deep learning (DL) offers automated, precise, and fine-grained quantitative indices for assessing postural control via video analysis.
• The clinical applicability of these DL-based tools needs further investigation.

Purpose of the Study:

• To compare the clinical applicability of three DL-based pose estimators for postural control assessment.
• To evaluate accuracy and processing speed of different DL pose estimation methods.
• To determine which quantitative indices derived from DL best correlate with occupational therapists' clinical evaluations.

Main Methods:

• Comparison of three deep-learning-based pose estimation techniques.
• Assessment of pose estimation accuracy and processing speed.
• Validation of DL-derived quantitative postural control indices against clinical judgment.

Main Results:

• The study evaluated the accuracy and processing speed of different DL pose estimators.
• It identified which DL-derived quantitative indices best reflected occupational therapists' clinical assessments.
• A DL framework demonstrated potential for more fine-grained postural control quantification.

Conclusions:

• Deep learning techniques offer a promising approach for enhanced, detailed quantification of postural control.
• This technology can potentially improve occupational therapy practice by providing objective, precise assessment tools.
• Further research is needed to fully integrate these advanced assessment tools into clinical settings.