Table 2.
Classification performances obtained by online CancerMath (CM) application and by training A, B, C, and D models on our dataset. The prediction performance of A, B, C, and D models obtained on the hold-out training set were evaluated on 100 ten-fold cross-validation rounds and summarized in terms of median, 1st, and 3rd quartile.
| Model | Performance Measure | Hold-Out Training Set | Hold-Out Test Set |
|---|---|---|---|
| CM on line | AUC (%) | 64.7 | 68.6 |
| Acc (%) | 68.3 | 66.2 | |
| Sens (%) | 46.4 | 41.5 | |
| Spec (%) | 73.6 | 75.2 | |
| CM features (A) | AUC (%) | 68.0 (67.6–68.3) | 68.6 |
| Acc (%) | 57.6 (55.4–66.2) | 51.5 | |
| Sens (%) | 72.3 (58.4–76.7) | 73.6 | |
| Spec (%) | 54.2 (50.5–67.9) | 43.4 | |
| CM features + Her2 (B) | AUC (%) | 67.6 (67.2–68.0) | 67.8 |
| Acc (%) | 56.4 (55.1–62.6) | 52.0 | |
| Sens (%) | 74.2 (62.9–77.7) | 73.6 | |
| Spec (%) | 52.1 (50.1–63.7) | 42.1 | |
| CM features + Ki67 (C) | AUC (%) | 67.4 (67.0–67.7) | 68.0 |
| Acc (%) | 56.4 (55.2–63.5) | 50.5 | |
| Sens (%) | 74.2 (61.6–76.5) | 69.8 | |
| Spec (%) | 52.0 (50.2–64.3) | 45.5 | |
| CM features + Ki67 + HER2 (D) | AUC (%) | 64.1 (63.8–64.6) | 65.4 |
| Acc (%) | 58.5 (56.1–61.3) | 53.8 | |
| Sens (%) | 68.1 (52.5–60.6) | 70.4 | |
| Spec (%) | 55.9 (63.2–71.9) | 48.3 |