Computational Pathology Classification on CPTAC k=10 (test)
92.9LUNG STPRISM
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| PRISMk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing, aggregation=mean over patch embeddings, multi-FM weighting=true2024.11 | 92.9 | 67.5 | 60.4 | 62 | 64.5 | 62.8 | 54.4 | 65.7 | 55.7 | 50.4 | 63.6 | |
| COBRA-V2k=10, Training Dataset=TCGA, Protocol=Few-shot linear probing, aggregation=mean over patch embeddings2024.11 | 81.6 | 64 | 61.9 | 59.7 | 59.2 | 56.5 | 56.2 | 64.6 | 59.2 | 51.3 | 61.4 | |
| COBRA-H0k=10, Training Dataset=TCGA, Protocol=Few-shot linear probing, aggregation=mean over patch embeddings2024.11 | 80.9 | 69.8 | 61.5 | 58.8 | 55.7 | 51.4 | 51.4 | 60.9 | 52.1 | 50.7 | 59.3 | |
| MADELEINEk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing, aggregation=mean over patch embeddings2024.11 | 80.6 | 59.2 | 57.4 | 57.1 | 60.2 | 55.2 | 54.3 | 56.6 | 50.6 | 51 | 58.2 | |
| COBRA-UNIk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing, aggregation=mean over patch embeddings2024.11 | 79.7 | 65.6 | 61 | 58.9 | 58.6 | 55.3 | 51.6 | 62.1 | 57 | 51.1 | 60.1 | |
| CONCHk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing2024.11 | 75.9 | 56 | 51.1 | 56 | 61.3 | 58.9 | 52.3 | 57.1 | 54.3 | 49.9 | 57.6 | |
| COBRA-CTPk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing, aggregation=mean over patch embeddings2024.11 | 74.3 | 59.9 | 58.4 | 56.1 | 58.2 | 55.4 | 52.2 | 58.4 | 52 | 49.5 | 57.4 | |
| CHIEFk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing, aggregation=mean over patch embeddings2024.11 | 68.9 | 59 | 56.1 | 54.2 | 62.3 | 59.7 | 52.3 | 58.1 | 50.4 | 50.2 | 57.1 | |
| GigaPathk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing2024.11 | 67.1 | 52.6 | 58.3 | 53.7 | 58.6 | 52.7 | 60.2 | 60 | 54.9 | 52.3 | 56.5 | |
| Virchow2k=10, Training Dataset=TCGA, Protocol=Few-shot linear probing2024.11 | 67.1 | 55.4 | 57.5 | 55.4 | 62.8 | 57.2 | 53.3 | 60 | 53.7 | 51.4 | 57.4 | |
| UNIk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing2024.11 | 64.9 | 53.6 | 57 | 55.6 | 58.2 | 58.6 | 51.9 | 57.7 | 51.3 | 53.3 | 56.2 | |
| H-Optimusk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing2024.11 | 64.7 | 58.2 | 55.2 | 52.2 | 57.4 | 52.8 | 51.1 | 57 | 50.3 | 51.2 | 55 | |
| GigaPath-SEk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing, aggregation=mean over patch embeddings2024.11 | 64.6 | 54.6 | 57.8 | 55.2 | 56.8 | 53.1 | 50.2 | 52.5 | 51.9 | 51.4 | 54.8 | |
| Virchowk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing2024.11 | 59.9 | 51.7 | 51.1 | 51.4 | 53.3 | 54.6 | 48.2 | 56 | 52.3 | 49.9 | 52.8 | |
| CTransPathk=10, Training Dataset=TCGA, Protocol=Few-shot linear probing2024.11 | 58.5 | 53.1 | 53 | 49.9 | 56.7 | 56.4 | 49.6 | 56.8 | 50.3 | 50.1 | 53.4 | |
| PRISMk=10, Training dataset=TCGA2024.11 | 0.929 | 0.675 | 0.604 | 0.62 | 0.645 | 0.628 | 0.544 | 0.657 | 0.557 | 0.504 | 0.636 | |
| COBRA-V2k=10, Training dataset=TCGA, Backbone=Virchow-2, Multi-FM weighting=true2024.11 | 0.816 | 0.64 | 0.619 | 0.597 | 0.592 | 0.565 | 0.562 | 0.646 | 0.592 | 0.513 | 0.614 | |
| COBRA-H0k=10, Training dataset=TCGA, Backbone=H-Optimus-0, Multi-FM weighting=true2024.11 | 0.809 | 0.698 | 0.615 | 0.588 | 0.557 | 0.514 | 0.514 | 0.609 | 0.521 | 0.507 | 0.593 | |
| MADELEINEk=10, Training dataset=TCGA2024.11 | 0.806 | 0.592 | 0.574 | 0.571 | 0.602 | 0.552 | 0.543 | 0.566 | 0.506 | 0.51 | 0.582 | |
| COBRA-UNIk=10, Training dataset=TCGA, Backbone=UNI, Multi-FM weighting=true2024.11 | 0.797 | 0.656 | 0.61 | 0.589 | 0.586 | 0.553 | 0.516 | 0.621 | 0.57 | 0.511 | 0.601 | |
| CONCHk=10, Training dataset=TCGA, Backbone=CONCH2024.11 | 0.759 | 0.56 | 0.541 | 0.56 | 0.613 | 0.589 | 0.523 | 0.571 | 0.543 | 0.499 | 0.576 | |
| COBRA-CTPk=10, Training dataset=TCGA, Backbone=CTransPath, Multi-FM weighting=true2024.11 | 0.743 | 0.599 | 0.584 | 0.561 | 0.582 | 0.554 | 0.522 | 0.584 | 0.52 | 0.495 | 0.574 | |
| CHIEFk=10, Training dataset=TCGA, Backbone=CHIEF2024.11 | 0.689 | 0.59 | 0.561 | 0.542 | 0.623 | 0.597 | 0.523 | 0.581 | 0.504 | 0.502 | 0.571 | |
| GigaPathk=10, Training dataset=TCGA, Backbone=GigaPath2024.11 | 0.671 | 0.526 | 0.583 | 0.537 | 0.586 | 0.541 | 0.527 | 0.602 | 0.549 | 0.523 | 0.565 | |
| Virchow2k=10, Training dataset=TCGA, Backbone=Virchow-22024.11 | 0.671 | 0.554 | 0.575 | 0.554 | 0.628 | 0.572 | 0.533 | 0.6 | 0.537 | 0.514 | 0.574 | |
| UNIk=10, Training dataset=TCGA, Backbone=UNI2024.11 | 0.649 | 0.536 | 0.57 | 0.556 | 0.582 | 0.586 | 0.519 | 0.577 | 0.513 | 0.533 | 0.562 | |
| H-Optimusk=10, Training dataset=TCGA, Backbone=H-Optimus-02024.11 | 0.647 | 0.582 | 0.552 | 0.522 | 0.574 | 0.528 | 0.511 | 0.57 | 0.503 | 0.512 | 0.55 | |
| GigaPath-SEk=10, Training dataset=TCGA, Backbone=Slide Encoder2024.11 | 0.646 | 0.546 | 0.578 | 0.552 | 0.568 | 0.531 | 0.502 | 0.525 | 0.519 | 0.514 | 0.548 | |
| Virchowk=10, Training dataset=TCGA, Backbone=Virchow2024.11 | 0.599 | 0.517 | 0.511 | 0.514 | 0.533 | 0.546 | 0.482 | 0.56 | 0.523 | 0.499 | 0.528 | |
| CTransPathk=10, Training dataset=TCGA, Backbone=CTransPath2024.11 | 0.585 | 0.531 | 0.53 | 0.499 | 0.567 | 0.564 | 0.496 | 0.568 | 0.503 | 0.501 | 0.534 |