Table 1.
AUROC of the best models for each task
| Best architecture | Per-patient AUROC | Per-tile AUROC | |
|---|---|---|---|
| Histology | Panoptes2 | 0.969 (0.905––1) | 0.870 (0.866–0.874) |
| CNV-H from endometrioid | Panoptes1 | 0.958 (0.886––1) | 0.864 (0.859–0.870) |
| CNV-H | Panoptes4 | 0.934 (0.851–1) | 0.731 (0.728–0.734) |
| POLE | Multi-model system | 0.890 (0.821–0.960) | 0.691 (0.683–0.700) |
| CNV-L | Panoptes1 | 0.889 (0.755–1) | 0.710 (0.705–0.716) |
| TP53 | Panoptes2 | 0.873 (0.768–0.977) | 0.713 (0.709–0.717) |
| FAT1 | Panoptes2 with clinical | 0.835 (0.666–1) | 0.639 (0.635–0.642) |
| MSI-high | InceptionResnetV1 | 0.827 (0.705–0.948) | 0.638 (0.635–0.641) |
| ZFHX3 | InceptionResnetV1 | 0.824 (0.689–0.959) | 0.637 (0.634–0.640) |
| PTEN | InceptionV2 | 0.781 (0.579–0.984) | 0.623 (0.620–0.627) |
| FGFR2 | Panoptes4 with clinical | 0.755 (0.540–0.970) | 0.550 (0.545–0.554) |
| MTOR | Panoptes1 | 0.724 (0.496–0.951) | 0.674 (0.670–0.678) |
| CTCF | Panoptes4 | 0.724 (0.518–0.931) | 0.571 (0.568–0.575) |
| PIK3R1 | InceptionResnetV1 | 0.702 (0.524–0.880) | 0.596 (0.593–0.599) |
| PIK3CA | Panoptes4 | 0.689 (0.532–0.847) | 0.526 (0.523–0.530) |
| ARID1A | InceptionResnetV2 | 0.683 (0.513–0.853) | 0.542 (0.538–0.545) |
| JAK1 | Panoptes2 with clinical | 0.662 (0.410–0.940) | 0.612 (0.605–0.618) |
| CTNNB1 | InceptionResnetV2 | 0.648 (0.439–0.858) | 0.619 (0.616–0.622) |
| KRAS | Panoptes2 with clinical | 0.638 (0.404–0.871) | 0.515 (0.510–0.519) |
| FBXW7 | InceptionV3 | 0.629 (0.366–0.892) | 0.606 (0.602–0.609) |
| RPL22 | InceptionV3 | 0.632 (0.395–0.868) | 0.517 (0.512–0.522) |
| BRCA2 | InceptionResnetV1 | 0.613 (0.318–0.908) | 0.624 (0.620–0.629) |
Bootstrapped 95% confidence intervals (CIs) are listed in parentheses. AUROCs > 0.75 are listed in bold.