Real-time and digital remote nutritional assessment framework with the use of smartphone-enabled facial morphometrics and machine learning- a proof of concept.

Abstract

Background: Current methods for assessing nutrition are often resource-intensive, requiring significant time, financial investment, and specialized equipment alongside clinical expertise.

Objective: This research introduces an innovative approach that emphasizes accessible, scalable, and efficient digital solutions by leveraging facial morphometrics and machine learning to predict essential nutritional indicators.

Methods: The cross-sectional observational study involved 71 free-living Chinese adults (30 males, 41 females) aged 50-85. Utilizing widely accessible smartphone technology, 3D facial scans were employed to forecast nutritional metrics. The predictive performance of two machine-learning models, Random Forest (RF) and Extreme Gradient Boosting (XGB), was evaluated through ten-fold stratified cross-validation.

Results: The RF model outperformed the XGB model, showing high predictive accuracy (median r² 0.51 to 0.92) for six parameters: muscle mass, basal metabolic rate (BMR), visceral fat index, appendicular skeletal muscle mass index, total body fat percentage, and hand grip strength. The highest predictive accuracy was found in muscle mass (r² = 0.92) and BMR (r² = 0.88) indicating strong correlations.

Conclusions: This non-invasive, economical technology presents a scalable approach to nutritional assessment with notable benefits for public health. The precise prediction of muscle mass and BMR facilitates efficient community-based screenings for undernutrition and frailty among older adults, while analysing body fat percentage aids in identifying overnutrition and related health risks. This digital approach shows significant potential for enhancing health outcomes on a population level through early detection and intervention.