Objective Falls Risk Assessment Using Markerless Motion Capture and Representational Machine Learning.

Area of Science:

• Gerontology
• Biomedical Engineering
• Computer Science

Background:

• Falls are a critical health concern for individuals over 65 globally.
• Objective fall risk assessment is underutilized in clinical settings, with common methods being time-intensive observational tests.
• Current fall risk assessment often relies on metrics like the time taken for the five times sit-to-stand test, which may not capture full movement dynamics.

Purpose of the Study:

• To develop and evaluate a computer-aided diagnostic tool for objective fall risk assessment.
• To leverage markerless motion capture and advanced modeling to gain deeper insights into movement patterns related to fall risk.
• To introduce a novel scoring system for continuous fall risk stratification.

Main Methods:

• Utilized markerless motion capture technology to track skeletal joint movements.
• Employed a Long Short-Term Memory (LSTM) autoencoder model to derive a distance measure from movement data.
• Developed a new scoring system based on the derived distance measure for continuous fall risk assessment.
• Evaluated the method on the KINECAL dataset, which includes recordings of 90 individuals performing 11 clinical movements.

Main Results:

• Achieved an accuracy of 0.84 in identifying individuals at elevated risk of falls.
• The developed scoring system allows for the placement of individuals on a continuous scale according to their fall risk.
• Demonstrated the potential for richer insights beyond simple time-based metrics.

Conclusions:

• The proposed markerless motion capture and LSTM autoencoder method offers an accurate and objective approach to fall risk assessment in older adults.
• The new continuous scoring system enhances the ability to differentiate fall risk levels.
• The method shows promise for clinical applications in fall prevention and rehabilitation, aligning with the KINECAL dataset's goals.