Table 2 iESPer scores for RCC and KTX
From: Ecologically sustainable benchmarking of AI models for histopathology
RCC Task (iESPer) | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
MODEL | CO2eq/Slide (g) | AUROC | 95%CI | ACCURACY | 95%CI | PRECISION | 95%CI | RECALL | 95%CI | F1 | 95%CI |
TransMIL | 0.046 | 0.964 | 0.936–0.986 | 0.8 | 0.732–0.868 | 0.736 | 0.634–0.835 | 0.695 | 0.577–0.822 | 0.676 | 0.556–0.798 |
CLAM | 0.048 | 0.937 | 0.908–0.960 | 0.815 | 0.753–0.874 | 0.833 | 0.764–0.897 | 0.54 | 0.442–0.661 | 0.583 | 0.453–0.717 |
InceptionV3 | 0.073 | 0.665 | 0.511–0.788 | 0.41 | 0.238–0.580 | 0.451 | 0.258–0.640 | 0.41 | 0.234–0.592 | 0.394 | 0.218–0.584 |
ViT | 0.065 | 0.693 | 0.593–0.784 | 0.643 | 0.571–0.718 | 0.448 | 0.270–0.654 | 0.393 | 0.316–0.485 | 0.37 | 0.272–0.484 |
Prov-GigaPath | 0.229 | 0.349 | 0.336–0.360 | 0.303 | 0.279–0.325 | 0.261 | 0.206–0.318 | 0.214 | 0.177–0.259 | 0.219 | 0.175–0.267 |
KTX Task (iESPer) | |||||||||||
MODEL | CO2eq/Slide (g) | AUROC | 95%CI | ACCURACY | 95%CI | PRECISION | 95%CI | RECALL | 95%CI | F1 | 95%CI |
TransMIL | 0.046 | 0.579 | 0.501–0.660 | 0.279 | 0.209–0.368 | 0.428 | 0.322–0.533 | 0.265 | 0.200–0.345 | 0.257 | 0.184–0.343 |
CLAM | 0.048 | 0.547 | 0.472–0.627 | 0.234 | 0.169–0.312 | 0.294 | 0.216–0.385 | 0.236 | 0.171–0.306 | 0.223 | 0.157–0.295 |
InceptionV3 | 0.073 | 0.391 | 0.334–0.451 | 0.062 | 0.038–0.097 | 0.007 | 0.004–0.011 | 0.093 | 0.093–0.093 | 0.017 | 0.012–0.024 |
ViT | 0.065 | 0.439 | 0.373–0.511 | 0.068 | 0.037–0.105 | 0.008 | 0.004–0.012 | 0.1 | 0.100–0.100 | 0.019 | 0.011–0.026 |
Prov-GigaPath | 0.229 | 0.188 | 0.160–0.222 | 0.065 | 0.044–0.089 | 0.110 | 0.062–0.168 | 0.055 | 0.041–0.072 | 0.042 | 0.025–0.062 |
