Radiomics-Based Muscle Quality Assessment from Multi-Site Multi-Vendor MRI.

Abstract

Muscle strength is critical for mobility, and its decline with aging contributes to frailty, falls, and loss of independence. Muscle cross-sectional area (CSA), which can be measured on MRI, only reflects muscle quantity and may not fully capture age-related changes in muscle composition, which significantly contribute to loss of muscle strength in older adults. We investigated strength prediction using multi-center/multi-scanner Dixon MRI from 141 adults across three cohorts: 82 healthy adults (Philips, 22-94 years), 24 healthy adults (GE, 30-79 years), and 35 older adults (Siemens, 70-91 years), including healthy, pre-frail, and sarcopenic participants. CSA was a strong predictor of muscle strength in the healthy cohorts, but radiomics features provided additional value when older adults were included. Fat-fraction radiomics showed the strongest and most consistent performance across cohort combinations, with the best results in the full combined cohort (R²=0.64 versus 0.42 for CSA alone). Water-only radiomics also retained predictive value, although generally less strongly than fat-fraction features. The most informative radiomics features reflected muscle size, signal magnitude, and internal heterogeneity, and SHAP analysis linked higher intramuscular heterogeneity to reduced strength. Taken together, these findings suggest that radiomics can capture muscle quality features that affect muscle force-generating capacity beyond size alone, particularly in older adults, and support automated MRI-based assessment of muscle composition and function, which potentially help future sarcopenia screening and functional risk stratification.