TABLE 4.
Sensitivity (%) at 2 and 4 false positives per image on DeepLesion’s three subsets of detection difficulty.
| Easy (FPs) | Medium (FPs) | Hard (FPs) | ||||
|---|---|---|---|---|---|---|
| 2 | 4 | 2 | 4 | 2 | 4 | |
| Faster R-CNN | 71.90 | 77.13 | 71.36 | 76.17 | 60.12 | 64.97 |
| Cascade R-CNN | 72.06 | 77.53 | 71.94 | 76.66 | 60.68 | 65.43 |
| YOLO-v3 | 70.10 | 75.78 | 69.27 | 74.84 | 59.27 | 63.99 |
| RetinaNet | 70.92 | 76.37 | 70.32 | 75.39 | 59.84 | 64.38 |
| 3DCE, 3 slices | 76.39 | 82.90 | 77.65 | 82.64 | 64.06 | 69.83 |
| FPN+3DCE, 3 slices | 78.65 | 83.87 | 79.21 | 84.56 | 65.07 | 71.39 |
| MVP-Net, 3 slices | 84.48 | 88.20 | 84.89 | 87.02 | 69.89 | 73.36 |
| Ours, 3 slices | 87.69 | 91.67 | 86.21 | 90.68 | 71.81 | 75.33 |