Frailty identification using a sensor-based upper-extremity function test: a deep learning approach.

Area of Science:

• Gerontology and Computational Health Science
• Biomechanical Engineering and Artificial Intelligence

Background:

• Aging populations necessitate improved frailty assessment tools.
• Current methods like Fried phenotype and Rockwood score have limitations.
• Efficient and accurate frailty prediction is crucial for preventing adverse health outcomes.

Purpose of the Study:

• To enhance frailty prediction accuracy in older adults.
• To investigate the utility of a combined biomechanical and deep learning approach.
• To identify key biomechanical predictors of frailty.

Main Methods:

• Recruited 312 older adults (non-frail, pre-frail, frail).
• Assessed frailty using Fried index, upper-extremity function (UEF) tests, and muscle force calculations.
• Employed machine learning (logistic regression, SVM) and deep learning (LSTM) models.

Main Results:

• Incorporating muscle model parameters significantly improved frailty prediction.
• The LSTM model achieved the highest accuracy (74%), outperforming SVM (67%) and regression (66%).
• Muscle co-contraction was a critical predictor, higher in frail individuals.

Conclusions:

• Integrating UEF tasks with deep learning offers superior frailty prediction.
• This approach can potentially serve as a robust and efficient clinical tool.
• Further validation in diverse populations is required to confirm generalizability.