Table 2 Validation ROC-AUC (binary tasks)/accuracy (non-binary tasks) values for all image-based classifiers.
From: Integrative multiomics-histopathology analysis for breast cancer classification
Image classifier validation set results | ||
|---|---|---|
Patient-level ROC-AUC/accuracy (95% CI) | Tile-level ROC-AUC/accuracy (95% CI) | |
Tumor vs. normal held-out test set | 0.985 (0.968–0.995) | 0.921 (0.919–0.924) |
Independent validation set | 0.950 (0.935–0.964) | 0.859 (0.857–0.860) |
Histological subtype held-out test set | 0.920 (0.845–0.993) | 0.800 (0.790–0.803) |
Independent validation set | 0.996 (0.984–1.0) | 0.843 (0.836–0.852) |
Estrogen receptor (ER) status +/− | 0.982 (0.97–0.994) | 0.929 (0.912–0.946) |
Progesterone receptor (PR) status +/− | 0.983 (0.977–0.989) | 0.908 (0.90–0.916) |
HER2 Receptor status +/− | 0.979 (0.971–0.981) | 0.829 (0.80–0.858) |
PAM50 Status (4 Class, Top-1) | 0.654 (0.636–0.672) | 0.406 (0.361–0.451) |
PAM50 Status (4 Class, Top-2) | 0.790 (0.763–0.817) | 0.609 (0.596–0.622) |
TP53 Mutation status | 0.833 (0.829–0.837) | 0.658 (0.634–0.682) |
