Table 1 Comparison of spinal CT models on the MultiSpine benchmark
Method | Segmentation | Identification | Lesion detection | ||
|---|---|---|---|---|---|
Dice (%) | ID Acc (%) | Precision (%) | Recall (%) | AP (%) | |
nnU-Net7 | 77.5 ± 0.4 | 60.7 ± 2.2 | 62.4 ± 1.5 | 58.2 ± 1.5 | 60.6 ± 1.5 |
UNETR19 | 88.1 ± 0.3 | 79.5 ± 1.0 | 65.8 ± 1.2 | 64.3 ± 1.2 | 65.8 ± 1.2 |
TransBTS46 | 88.3 ± 0.3 | 80.1 ± 1.0 | 66.4 ± 1.2 | 64.8 ± 1.2 | 66.3 ± 1.2 |
H2Former22 | 88.6 ± 0.3 | 81.3 ± 0.9 | 66.9 ± 1.2 | 65.7 ± 1.2 | 66.9 ± 1.2 |
Scribformer23 | 88.7 ± 0.3 | 82.4 ± 0.9 | 67.4 ± 1.1 | 68.2 ± 1.1 | 67.8 ± 1.1 |
BLDS47 | 84.4 ± 0.4 | 70.5 ± 1.6 | 60.8 ± 1.7 | 55.3 ± 1.7 | 57.4 ± 1.7 |
Dense-U-Net6 | 82.2 ± 0.4 | 75.6 ± 1.3 | 58.7 ± 1.9 | 52.9 ±1.9 | 55.4 ± 1.9 |
VerFormer48 | 83.9 ± 0.3 | 82.4 ± 0.9 | 66.3 ± 1.1 | 67.4 ± 1.1 | 67.2 ± 1.1 |
Tao et al.49 | 84.5 ± 0.3 | 83.3 ± 0.8 | 65.9 ± 1.2 | 66.8 ± 1.2 | 66.2 ± 1.2 |
VerteFormer50 | 86.0 ± 0.3 | 84.2 ± 0.7 | 67.9 ± 1.1 | 67.4 ± 1.1 | 67.7 ± 1.1 |
VertDetect51 | 80.5 ± 0.6 | 80.6 ± 1.0 | 58.6 ± 1.9 | 52.3 ± 1.9 | 55.2 ± 1.9 |
Spineclue52 | 80.3 ± 0.6 | 80.4 ± 1.0 | 60.4 ± 1.7 | 55.6 ± 1.7 | 57.3 ± 1.7 |
Ortho2D53 | 78.4 ± 0.7 | 78.5 ± 1.1 | 55.4 ± 2.0 | 50.7 ± 2.0 | 52.6 ± 2.0 |
VertNet54 | 82.3 ± 0.5 | 82.6 ± 0.8 | 60.5 ± 1.6 | 58.2 ± 1.6 | 59.5 ± 1.6 |
VertebraFormer (ours) | 89.3 ± 0.2 | 85.6 ± 0.6 | 71.1 ± 1.0 | 69.0 ± 0.9 | 68.7 ± 1.1 |
