Table 3.
Glioma classification results. The proposed EM-CNN-LR method achieved the best result, close to interobserver agreement between pathologists. (Sec. 5.4).
| Methods | Acc | mAP |
|---|---|---|
| CNN-Vote | 0.710 | 0.812 |
| CNN-SMI | 0.710 | 0.822 |
| CNN-Fea-SVM | 0.688 | 0.790 |
| EM-CNN-Vote | 0.733 | 0.837 |
| EM-CNN-SMI | 0.719 | 0.823 |
| EM-CNN-Fea-SVM | 0.686 | 0.790 |
| EM-Finetune-CNN-Vote | 0.719 | 0.817 |
| EM-Finetune-CNN-SMI | 0.638 | 0.758 |
| CNN-LR | 0.752 | 0.847 |
| CNN-SVM | 0.697 | 0.791 |
| EM-CNN-LR | 0.771 | 0.845 |
| EM-CNN-LR w/o spatial smoothing | 0.745 | 0.832 |
| EM-CNN-SVM | 0.730 | 0.818 |
| EM-Finetune-CNN-LR | 0.721 | 0.822 |
| EM-Finetune-CNN-SVM | 0.738 | 0.828 |
| SMI-CNN-SMI | 0.683 | 0.765 |
| NM-LBP | 0.629 | 0.734 |
| Pretrained CNN-Fea-SVM | 0.733 | 0.837 |
| Pretrained-CNN-Bow-SVM | 0.667 | 0.756 |
| Chance | 0.513 | 0.689 |