Using a Stacked Autoencoder for Mobility and Fall Risk Assessment via Time-Frequency Representations of the Timed Up...

Area of Science:

• Gerontology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Fall risk assessment is critical for aging populations, with mobility impairment being a key factor.
• Current continuous fall risk monitoring methods are resource-intensive.
• Remote, expert-free monitoring is needed for long-term elderly health surveillance.

Purpose of the Study:

• To develop a remote, long-term health monitoring method for elderly mobility impairment and fall risk.
• To assess fall risk without requiring specialized healthcare professionals.
• To leverage advanced signal processing and machine learning for accessible health monitoring.

Main Methods:

• Utilized time-frequency analysis (TFA) to transform triaxial accelerometer time-series data into image representations.
• Employed a stacked autoencoder (SAE), a deep neural network (DNN), for semi-supervised learning.
• Assessed mobility and fall risk based on the criteria of the Timed Up and Go test (TUG).

Main Results:

• The semi-supervised SAE model achieved high predictive accuracies: 89.1% (vertical), 93.4% (mediolateral), and 94.1% (anteroposterior) axes.
• TFA successfully generated richer image information from accelerometer signals.
• Demonstrated the effectiveness of deep learning in analyzing acceleration data for fall risk assessment.

Conclusions:

• Deep learning models, specifically SAEs, are effective for analyzing triaxial acceleration data.
• The developed method shows significant applicability for assessing elderly mobility and fall risk.
• This approach offers a viable solution for remote, continuous, and non-expert-dependent health monitoring.