Frailty Assessment Using Temporal Gait Characteristics and a Long Short-Term Memory Network

Area of Science:

• Gerontology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Frailty is a growing health concern in the aging global population.
• Current frailty assessment methods can be resource-intensive.
• Objective and accessible frailty detection is needed.

Purpose of the Study:

• To explore the use of temporal gait characteristics and Long Short-Term Memory (LSTM) networks for frailty assessment.
• To determine the feasibility of a machine learning approach for classifying elderly individuals into robust, pre-frail, and frail categories.

Main Methods:

• Seventy-four community-dwelling elderly individuals participated.
• Gyroscopic data from foot sensors captured temporal gait parameters during a 7-meter walk.
• Gait sequence features and demographic data were used to train a Random Forest model incorporating an LSTM classifier.

Main Results:

• The Random Forest model achieved an F1-score of 0.931 in classifying frailty levels.
• Key inputs included age, sex, and LSTM-derived features from gait parameters like double-limb support, step, and stride times.
• The model demonstrated high accuracy in distinguishing between robust, pre-frail, and frail participants.

Conclusions:

• Combining temporal gait analysis with LSTM networks offers a novel and effective method for frailty assessment.
• This approach holds promise for developing self-monitoring tools for early frailty detection at home.
• Facilitating timely medical intervention to manage frailty in the elderly population.