Table 2 Comparison of segmentation results of testing dataset with different margins
From: Weakly-supervised convolutional neural networks of renal tumor segmentation in abdominal CTA images
| Â | Â | DSC | HD | ASD |
|---|---|---|---|---|
d = 0 | k = 0 (WCE Loss) | 0.788 | 65.806 | 6.265 |
k = 3 (WCE Loss) | 0.822 | 34.187 | 3.889 | |
k = 3 (VWCE Loss) | 0.834 | 40.617 | 3.361 | |
k = 3 (VWCE Loss) + 3D MC | 0.834 | 14.346 | 2.664 | |
d = 5 | k = 0 (WCE Loss) | 0.733 | 32.459 | 5.332 |
k = 3 (WCE Loss) | 0.784 | 70.948 | 7.988 | |
k = 3 (VWCE Loss) | 0.820 | 37.633 | 3.879 | |
k = 3 (VWCE Loss) + 3D MC | 0.826 | 15.811 | 2.838 | |
d = 10 | k = 0 (WCE Loss) | 0.695 | 58.286 | 7.499 |
k = 3 (WCE Loss) | 0.720 | 81.611 | 7.804 | |
k = 3 (VWCE Loss) | 0.741 | 36.127 | 4.672 | |
k = 3 (VWCE Loss) + 3D MC | 0.742 | 21.233 | 4.350 |