Listening to muscles: a machine learning approach to low-cost sarcopenia detection for older adults

Sarcopenia is characterized by progressive declines in skeletal muscle mass, strength, and quality, yet accessible and standardized methods for assessing intrinsic muscle quality in community settings remain limited. Advanced imaging and medical-grade instruments are costly and impractical for large-scale screening. We developed a low-cost, portable, and noninvasive acoustic device based on myophonogram principles to quantify muscle quality using machine learning regression models.

In this community-based study, individuals aged 65 years or older were recruited from five community centers. Acoustic signals from the gastrocnemius muscle were recorded and decomposed into 52 time- and frequency-domain features. Four ensemble regression models-gradient boosting, extreme gradient boosting, histogram-based gradient boosting, and light gradient boosting machine-were trained to estimate muscle parameters. Reference standards included ankle dorsiflexion strength and ankle plantar-flexion strength measured by MicroFET dynamometry, as well as appendicular skeletal muscle mass index measured by bioelectrical impedance analysis RESULTS: A total of 160 leg measurements were used for model development, and an independent cohort of 16 participants was included for external validation. All models demonstrated strong predictive performance. Histogram-based gradient boosting, light gradient boosting machine, and extreme gradient boosting achieved the highest accuracy for predicting ankle dorsiflexion strength (R² = 0.91), ankle plantar-flexion strength (R² = 0.86), and appendicular skeletal muscle mass index (R² = 0.82), respectively. Ten-fold cross-validation confirmed model robustness, and external validation showed low absolute percentage errors.

This noninvasive acoustic device enables accurate estimation of surrogate measures of muscle quality - namely ankle dorsiflexion strength, ankle plantar-flexion strength, and appendicular skeletal muscle mass index - and offers a scalable point-of-care approach for community-based screening of individuals at risk of sarcopenia.