Table 3.
Simple and multiple logistic regression models to measure the predictive power of percentage mammographic density (PMD) and selected features within and across the five texture feature groups, via feature group scores and the final feature score, respectivelya
| Training data set | Validation data set | ||||||
|---|---|---|---|---|---|---|---|
| Unadjusted | Adjusted for age and BMI | Adjusted for age, BMI, parity, family history, and age at FTP | |||||
| Texture features included | AUC | AUC | OR (95% CI) | AUC | OR (95% CI) | AUC | OR (95% CI) |
| Noneb | - | - | - | 0.60 | - | 0.65 | - |
| PMD | 0.53 | 0.51 | 1.05 (0.89-1.23) | 0.61 | 1.24 (1.00-1.55) | 0.66 | 1.19 (0.93-1.53) |
| Moment-based features (n = 8, group 1) |
0.66 | 0.58 | 1.46 (1.22-1.73) | 0.62 | 1.43 (1.19-1.72) | 0.67 | 1.41 (1.14-1.75) |
| Form-based features (n = 16) | 0.67 | 0.59 | 1.47 (1.23-1.74) | 0.64 | 1.44 (1.20-1.74) | 0.67 | 1.49 (1.21-1.84) |
| Statistical features (n = 46) |
0.82 | 0.72 | 2.40 (1.98-2.90) | 0.73 | 2.28 (1.87-2.78) | 0.74 | 2.36 (1.88-2.96) |
| Structural features (n = 23) |
0.77 | 0.65 | 1.64 (1.38-1.95) | 0.68 | 1.60 (1.34-1.92) | 0.71 | 1.70 (1.39-2.08) |
| Spectral features (n = 6) |
0.71 | 0.65 | 1.67 (1.40-1.99) | 0.67 | 1.57 (1.30-1.90) | 0.68 | 1.60 (1.29-1.98) |
| Selected features across all feature groups (final model; n = 46) | 0.85 | 0.75 | 2.65 (2.18-3.21) | 0.75 | 2.55 (2.08-3.11) | 0.79 | 2.88 (2.28-3.65) |
| Selected features across all feature groups + PMD | 0.85 | 0.75 | 2.63 (2.17-3.18) | 0.75 | 2.52 (2.06-3.08) | 0.79 | 2.86 (2.26-3.62) |
The area under the curve (AUC) of the regression models and the odds ratio (OR) per standard-deviation (SD) change for the feature scores with 95% confidence intervals are shown. Features were selected as described in the Patients and Methods sections.
AUC, area under the curve; BMI, body mass index; CI, confidence interval; FFTP, first term pregnancy; OR, odds ratio; PMD, percentage mammographic density; SD, standard deviation. a In the training data, each logistic regression model used selected features as independent variables; in validation data, the logistic regression models used the feature group scores and the final feature score, respectively, as independent variable. Adjusted analyses with regular risk factors as additional independent variables. bPrediction only with regular risk factors.