Deep Learning-Based Sarcopenia Classification through Gait Video Analysis with a Single Mobile Camera.

Area of Science:

• Gerontology
• Biomedical Engineering
• Computer Science

Background:

• Global population aging necessitates effective strategies for healthy aging.
• Sarcopenia, an age-related decline in muscle mass and strength, impacts elderly health.
• Current sarcopenia detection methods are often costly and require expert intervention.

Purpose of the Study:

• To develop and validate a novel vision-based approach for sarcopenia identification using gait analysis.
• To provide an accessible and practical method for early sarcopenia detection and monitoring.
• To reduce reliance on specialized equipment and expertise for sarcopenia assessment.

Main Methods:

• Recruited 92 elderly individuals (60 with sarcopenia, 32 healthy controls).
• Captured walking motion using digital cameras to extract 2D skeleton sequences.
• Utilized a deep learning model trained on skeleton sequences and gait features for classification.

Main Results:

• The deep learning model achieved 82.88% sample-wise accuracy and 94.44% subject-wise accuracy in classifying sarcopenia.
• Demonstrated the efficacy of gait analysis via 2D skeleton sequences for sarcopenia detection.
• Validated the approach on an independent test dataset.

Conclusions:

• Vision-based gait analysis offers a significant advancement for sarcopenia detection and monitoring.
• This method presents a promising, accessible solution for at-home muscle health management.
• Eliminates the need for specialized equipment and expert interpretation in sarcopenia assessment.