Table 3.

Differentiation between patients with and without osteoporotic vertebral fractures including texture analysis – analysis on patient level.

Term	Description	β coefficient	95%-CI	p-value
Integral vBMD	–	-0.669	-0.010;-0.005	<0.001
CT_SRE	Higher-order texture feature, representing the short-run emphasis	0.721	154.622;287.516	<0.001
CT_Varianceglobal	Global texture feature, representing the spread of gray-level distribution	-0.519	-0.021;-0.008	<0.001
PDFF_Variance	Second-order texture feature, representing the voxel co-occurrence distribution	0.351	5.390;36.408	0.011

This table shows the variables kept in the final linear regression model (adjusted R² [R² _a] = 0.81 (F(6, 19) = 19.2, p < 0.001) after a stepwise approach using the binary fracture status (at least one osteoporotic vertebral fracture present/no osteoporotic vertebral fracture present) as the dependent variable (analyses on patient level). Specifically, it included integral volumetric bone mineral density (vBMD) and the texture features CT_SRE, CT_Varianceglobal, and PDFF_Variance (β coefficients, 95%-confidence intervals [CIs], and p-values shown per texture feature). Patient age, sex, and the number of independent variables were considered for adjustment. For analyses on patient level, integral and trabecular vBMD, PDFF, T2*, and texture features were averaged over the included vertebral bodies to provide one value per parameter in each patient, respectively.