Skeletal Muscle Mass Estimation from Lower Leg Digital Images Using Machine Learning.

OBJECTIVES: In the current study, a convolutional neural network (CNN) model for estimating the skeletal muscle index (SMI) from lower leg digital images was developed, and its predictive performance was evaluated. METHODS: One hundred healthy adults participated in the study; 50 men and 50 women (median age 37.5 years, interquartile range 28-44 years). Digital images of the non-dominant lower leg were obtained from lateral and posterior views. A CNN model was trained to estimate SMI after applying data augmentation and Canny edge detection. Predictive accuracy was evaluated using fivefold cross-validation (80% training, 20% testing). RESULTS: The mean SMI was 7.0 ± 1.1 kg/m2. The mean absolute percentage error ranged from 3.70% to 4.15% for lateral images and from 4.23% to 4.85% for posterior images. The respective concordance correlation coefficients ranged from 0.93 to 0.94 and from 0.90 to 0.91, indicating strong agreement with the reference values. CONCLUSIONS: The CNN model accurately estimated SMI from lower leg digital images, with prediction errors below 5%. This simple, noninvasive method could serve as a practical tool for screening skeletal muscle mass.