Table 1 Comparison of WSI classification and location performance between CAMCSA and other methods on the Camelyon16 dataset. CAMCSA/without CSA denotes the experimental results on Camelyon16 of CAMCSA before introducing the CSA module.
From: Cross-slide augmentation for whole slide image classification based on class activation map
Methods | Camelyon16 | ||
|---|---|---|---|
ACC | AUC | FROC | |
Mean pooling | 0.7820 | 0.7917 | 0.1171 |
Max pooling | 0.8178 | 0.8165 | 0.3421 |
ABMIL | 0.8643 | 0.8823 | 0.4042 |
DSMIL | 0.8996 | 0.9137 | 0.4213 |
Trans-MIL | 0.8879 | 0.9221 | 0.4409 |
DTFD-MIL | 0.9029 | 0.9502 | 0.4587 |
MHIM-MIL | 0.9201 | 0.9608 | 0.4590 |
CAMCSA/without CAM and CSA | 0.9021 | 0.9343 | 0.4378 |
CAMCSA/without CAM | 0.9271 | 0.9532 | 0.4652 |
CAMCSA/without CSA | 0.9399 | 0.9676 | 0.4742 |
CAMCSA | 0.9433 | 0.9793 | 0.4808 |
