Table 2.
Summary of the datasets used for training and testing the convolutional neural network.
| Cancer type | Number of WSIs used for network training | Regional classification | ||
|---|---|---|---|---|
| Cutaneous melanoma | Total | 27 | 6 categories | Superpixels for training |
| Training | 22 | Tumor tissue | 21940 | |
| Testing | 5 | Stroma | 12419 | |
| Normal epidermis | 1646 | |||
| Lymphocytes cluster | 2367 | |||
| Fat | 15484 | |||
| Empty/white space | 3412 | |||
| Triple-negative breast cancer | Total | 23 | 6 categories | Superpixels for training |
| Training | 18 | Tumor tissue | 18873 | |
| Testing | 5 | Stroma | 24220 | |
| Necrosis | 15102 | |||
| Lymphocytes cluster | 3472 | |||
| Fat | 10044 | |||
| Empty/white space | 16473 | |||
| High-risk neuroblastoma (mouse model) | Total | 60 | 8 categories | Superpixels for training |
| Training | 44 | Region of undifferentiated neuroblasts | 20512 | |
| Testing | 16 | Tissue damage (necrosis/apoptosis) | 17645 | |
| Differentiation region | 5740 | |||
| Lymphocytes cluster | 4009 | |||
| Hemorrhage (blood) | 6124 | |||
| Muscle | 6415 | |||
| Kidney | 14976 | |||
| Empty/white space | 21470 | |||
Note that the testing datasets consisted of whole-slide images from different patients from the training dataset.