CT radiomic features detect mineral, collagen, and water contributions to cortical bone texture

Without knowing which compositional properties drive specific features, radiomics cannot move from empirical association toward mechanistic interpretation. This gap is particularly relevant in several metabolic bone diseases where collagen and water abnormalities contribute to skeletal fragility despite sometimes preserved bone mineral density. Ten cortical cross-sections from a single fresh human cadaveric femur underwent stepwise selective removal of mineral (hydrochloric acid), collagen (sodium hypochlorite), and free and bound water (thermal treatment). Specimens were imaged by high-resolution peripheral quantitative computed tomography (HR-pQCT) at baseline, halfway, and full removal. Compositional loss was confirmed by dual-energy X-ray absorptiometry and thermogravimetric analysis. Radiomic features were selected through variance filtering, correlation thresholding, and bootstrap stability ranking, with group comparisons by one-way ANOVA or Kruskal-Wallis test. Each compositional component produced distinct, reproducible radiomic signatures. Mineral removal drove the largest radiodensity change, with Maximum intensity decreasing 13.5% at full demineralization accompanied by a greater than twofold increase in Contrast. Collagen removal generated biphasic responses in surface geometry, with Mesh Surface expanding 2.8% at halfway removal before contracting below baseline at full removal. Despite CT’s limited sensitivity to organic matrix, both water compartments produced measurable texture changes: free water removal progressively narrowed grayscale Range by 6.4%, while bound water removal produced distinct shifts in spatial Complexity and intensity Range. CT radiomic features are sensitive to mineral, collagen, and water contributions to cortical bone texture, including components conventionally considered invisible to density-based imaging. These preliminary findings suggest that individual cortical bone matrix components may produce distinct radiomic feature behaviors in a controlled ex vivo model, providing an initial compositional basis for interpreting bone texture in HR-pQCT images and informing the development of composition-aware imaging biomarkers for metabolic bone diseases.