TABLE II. The Segmentation and Classification Performance of Our Model, Human Expert and Other Models. The Metrics are Presented in the Format of Mean (Standard Deviation). Note That the 3D Lesion Segmentation and Classification Subnets in Our Model Are Denoted as Seg. and Cls., Respectively.
| Modality | Time (s) | DSC (%) | Sensitivity (%) | Specificity (%) | NSD (%) | RMSD (mm) | ||
|---|---|---|---|---|---|---|---|---|
| Segmentation | Human expert | 3D | 869.2 | 60.3 (10.8) | 61.2 (7.2) | 88.7 (1.2) | 60.2 (3.6) | 12.8 (4.1) |
| U-Net++L1 | 2D | 1.0 | 61.2 (5.7) | 66.7 (6.4) | 90.9 (0.8) | 62.4 (3.6) | 13.6 (2.3) | |
| U-Net++L4 | 2D | 8.2 | 65.3 (6.2) | 71.3 (5.8) | 92.8 (0.6) | 66.8 (3.1) | 9.6 (2.6) | |
| DenseVNet | 3D | 2.8 | 63.7 (7.4) | 69.2 (5.9) | 92.1 (0.7) | 63.6 (2.7) | 7.8 (3.1) | |
| COPLE-Net | 2D | 10.8 | 67.2 (6.8) | 73.4 (7.2) | 93.2 (0.6) | 68.3 (1.9) | 5.2 (2.2) | |
| FSS-2019-nCov | 2D | 8.7 | 67.9(7.6) | 74.1 (5.2) | 93.8 (0.7) | 68.8 (2.1) | 5.4 (3.0) | |
| DeepSC-COVID w/o Cls. | 3D | 1.1 | 66.2 (7.7) | 72.7 (4.9) | 92.7 (0.8) | 64.3 (2.8) | 6.2 (2.5) | |
| DeepSC-COVID | 3D | 2.2 | 73.3 (8.5) | 80.2 (6.8) | 95.6 (0.7) | 71.8 (2.6) | 2.8 (1.6) | |
| Modality | Time (s) | Accuracy (%) | Sensitivity (%) | Specificity (%) | F1-score (%) | AUC (%) | ||
| Classification | Human expert | 3D | 378.7 | 95.8 (0.4) | 96.0 (0.4) | 97.9 (0.7) | 95.7 (0.3) | − |
| ResNet-50 | 2D | 9.2 | 77.7 (0.7) | 79.2 (0.2) | 87.7 (0.6) | 77.9 (0.8) | 92.0 (0.1) | |
| 3D ResNet-50 | 3D | 0.6 | 82.5 (0.2) | 82.5 (0.7) | 90.6 (0.2) | 82.0 (0.5) | 94.5 (0.2) | |
| COVNet | 2D | 5.3 | 87.2 (0.8) | 87.6 (0.3) | 93.7 (0.4) | 87.5 (0.3) | 97.4 (0.1) | |
| DeCoVNet | 3D | 2.3 | 89.2 (0.5) | 89.0 (0.4) | 94.3 (0.2) | 88.9 (0.8) | 98.3 (0.1) | |
| DeepSC-COVID w/o Seg. | 3D | 0.7 | 85.2(0.4) | 85.3 (0.2) | 91.9 (0.3) | 84.9 (0.1) | 94.4 (0.2) | |
| DeepSC-COVID | 3D | 2.2 | 94.5 (0.3) | 94.7 (0.2) | 97.3 (0.5) | 94.2 (0.6) | 98.8 (0.1) | |