Research Article
SC-Dynamic R-CNN: A Self-Calibrated Dynamic R-CNN Model for Lung Cancer Lesion Detection
Table 1
Comparisons with different models on our lung cancer dataset.
| | Method | Backbone | Adenocarcinoma | Squamous | Small cell | mAP | | AP50 | AP75 | AP50 | AP75 | AP50 | AP75 |
| | RetiNanet [21] | ResNet-50 | 87.7% | 67.8% | 89.7% | 79.9% | 88.1% | 77.8% | 81.8% | | SSD [38] | ResNet-50 | 80.7% | 61.4% | 89.2% | 78.4% | 86.2% | 77.3% | 78.9% | | Faster R-CNN [23] | ResNet-50 | 81.6% | 62.5% | 90.5% | 80.3% | 89.7% | 79.4% | 80.1% | | Libra R-CNN [27] | ResNet-50 | 81.9% | 71.4% | 89.9% | 81.5% | 89.3% | 83.2% | 82.9% | | Cascade R-CNN [25] | ResNet-50 | 82.7% | 73.5% | 90.1% | 82.9% | 90.1% | 84.9% | 84.0% | | SC-Dynamic R-CNN | ResNet-50 | 91.6% | 77.3% | 91.5% | 88.2% | 91.4% | 88.6% | 88.1% |
|
|
