Association of radiomic features of skeletal muscle on CT images with muscle function and physical performance in older men.
Authors
Affiliations (8)
Affiliations (8)
- California Pacific Medical Center Research Institute, San Francisco, CA, USA.
- Department of Radiology, Stanford University School of Medicine, Palo Alto, CA, USA.
- Department of Biomedical Engineering, Wake Forest University School of Medicine, Winston-Salem, NC, USA.
- Department of Geriatric Medicine, Primary Care & Population Health, Stanford University School of Medicine, Palo Alto, CA, USA.
- Geriatric Research Clinical and Education Center, Veterans Affairs Health System, Palo Alto, CA, USA.
- Department of Medicine, University of California Davis, Davis, CA, USA.
- Department of Endocrinology, Diabetes and Clinical Nutrition, Oregon Health & Science University, Portland, OR, USA.
- Department of Epidemiology and Biostatistics, University of California San Francisco, San Francisco, CA, USA.
Abstract
Machine learning applied to computed tomography (CT) images captures variations in skeletal muscle texture and structure not detectable by conventional measures. These novel 'radiomic' features may offer added value in predicting muscle function and physical performance beyond traditional CT-derived muscle area and density. We aimed to identify radiomic features of skeletal muscle associated with grip strength, leg power and walking speed in older men. In the Osteoporotic Fractures in Men study (n = 3404; 73.8 ± 5.9 years), participants underwent baseline CT scans (trunk L1, L3; right and left thigh) and assessments of grip strength, 6 m walk and leg power (Nottingham Power Rig). Muscle area and density were derived from automatically segmented CT images. Radiomic features were extracted using PyRadiomics. Elastic net regression and factor analysis identified key radiomic features; associations with muscle function/performance were assessed using regression models. Factor analysis identified nine factors for Trunk-L1 and eight for the other regions. Trunk-based factors significantly improved model fit for leg power, grip strength and walking speed (P < .05). Factor 1, representing body size and muscle texture complexity, was the most consistent predictor across outcomes. The Gray-Level Co-occurrence Matrix feature 'cluster prominence' was inversely associated with walking speed (β = -0.06 at L1; -0.05 at L3) and leg power (β = -0.05 at L1), independent of age, height, weight, muscle CSA, muscle density and technical group. CT-derived radiomic features in the trunk region may reflect skeletal muscle structural characteristics that independently relate to strength, power and mobility in older men.