Machine Learning Based Walking Aid Detection in Timed Up-and-Go Test Recordings of Elderly Patients

Area of Science:

• Gerontology
• Biomedical Engineering
• Clinical Biomechanics

Background:

• Frailty and falls are significant contributors to morbidity and disability in the elderly population.
• The Timed Up-and-Go (TUG) test is a recognized tool for assessing fall risk in older adults.
• Identifying factors that predict falls, such as walking aid use, is crucial for effective intervention.

Purpose of the Study:

• To analyze the potential of the TUG test for predicting falls in elderly individuals.
• To develop a method for predicting walking aid use based on ultrasonic TUG test recordings.
• To evaluate the accuracy of the developed prediction model.

Main Methods:

• A clinical study was conducted with 138 participants aged 65 years and older living in nursing homes.
• Ultrasonic recordings of the TUG test were collected, noting whether participants used a walking aid.
• A Random Forest Classifier was employed to predict walking aid use from TUG test data.

Main Results:

• The Random Forest Classifier achieved a high accuracy in predicting walking aid use.
• The Area Under the Curve (AUC) reached 96.9% using a 20% leave-out evaluation strategy.
• The model demonstrated strong performance in distinguishing TUG recordings associated with walking aid use.

Conclusions:

• Automated data collection from TUG recordings can reliably predict walking aid use.
• This approach shows promise for enhancing the accuracy and utility of fall risk assessment tools.
• Integrating such predictive methods may lead to improved fall prevention strategies for the elderly.