Table 2.
The results of six state-of-the-art architectures of deep learning algorithms on various datasets using ARC. The numbers are measured based on TNu and FNu and represent accuracy percentages. The bold fonts indicate the best classification accuracies on datasets.
| Algorithms | Tumor tissue discrimination | Bladder biomarker discrimination | Breast biomarker discrimination | Lung tumor subtype discrimination (TMAD images) |
|---|---|---|---|---|
| CNN-basic | 100% | 71.5% | 79.2% | 73% |
| Inception V3-Last layer-4000 steps | 99.3% | 86% | 75.6% | 80% |
| Inception V3-Last layer-12000 steps | 99.3% | 87.5% | 76.8% | 78% |
| Inception V1-Fine tune | 100% | 99% | 90% | 92% |
| Inception-ResNet V2-Last layer | 96.6% | 85.5% | 78.4% | 75% |
| Inception V3-Fine tune | 100% | 98% | 90.8% | 90% |
| Algorithms | Lung tumor subtype discrimination (TCGA intra-images) | Lung tumor subtype discrimination (TCGA inter-images) | Score discrimination in bladder | Score discrimination in breast |
| CNN-basic | 68% | 61% | 47% | 40.5% |
| Inception V3-Last layer-4000 steps | 84% | 70% | 64.5% | 61% |
| Inception V3-Last layer-12000 steps | 80% | 70% | 64.5% | 59.5% |
| Inception V1-Fine tune | 100% | 83% | 77% | 56% |
| Inception-ResNet V2-Last layer | 84% | 66% | 58% | 45.5% |
| Inception V3-Fine tune | 100% | 79% | 76% | 56% |