Table 4.
Diagnostic Performance of Radiomic Models on Ki-67 level Prediction.
| Data Group | AUROC | Accuracy | Sensitivity | Specificity | |
|---|---|---|---|---|---|
| Grayscale | Training | 0.868 [0.775–0.947] | 0.843 [0.771–0.914] | 0.860 [0.778–0.931] | 0.769 [0.556–0.944] |
| Testing | 0.712 [0.528–0.875] | 0.600 [0.567–0.833] | 0.710 [0.615–0.893] | 0.600 [0.000–0.800] | |
| Kupffer CEUS | Training | 0.873 [0.763–0.960] | 0.786 [0.700–0.871] | 0.754 [0.655–0.847] | 0.923 [0.778–1.000] |
| Testing | 0.768 [0.621–0.896] | 0.700 [0.600–0.867] | 0.746 [0.667–0.920] | 0.800 [0.500–0.833] | |
| Difference p | Training | 0.873 | 0.12 | 0.069 | <0.001 |
| Testing | 0.048 | 0.2 | 0.834 | <0.001 |
96% Confidence Intervals were included in parenthesis. p-value was calculated with the DeLong test for AUROC comparison and a two-sided permutation test with 10,000 random resamples.