Table 2.
Association between texture features and case vs. control status and classification performance
| Region | Initial | 12–18-Month change | ||||
|---|---|---|---|---|---|---|
| Odds ratio (95% CI)a | Most important featuresb | c-statistic (95% CI) | Odds ratio (95% CI)a | Most important featuresb | c-statistic (95% CI) | |
| Tibia | 1.43 (0.99, 2.09) | Gr. variance, variance | 0.58 (0.58, 0.58) | 2.31 (1.42, 4.12)*** | Gr. mean, Gr. variance, contrast | 0.65 (0.63, 0.69)** |
| Femur | 1.62 (1.12, 2.44)** | Mean, variance, ASM | 0.60 (0.59, 0.60)* | 1.80 (1.17, 2.92)** | ASM, contrast, variance | 0.63 (0.61, 0.65)** |
| Combined | 1.84 (1.25, 2.80)** | F variance, F mean, T variance | 0.64 (0.64, 0.65)** | 3.76 (2.04, 7.82)*** | T Gr. variance, F ASM, T entropy | 0.68 (0.68, 0.68)*** |
T – tibial feature, F – femoral feature, Gr – gradient, ASM – angular second moment
* p < 0.05, ** p < 0.01, ***p < 0.001
aOdds ratio of being a progressor for each o standard deviation increase in texture score
bThree texture features with largest standardised coefficients (β) in the logistic regression model (note initial tibial model only included two texture features)