Comparisona of StackGen-Net with variants of DeepUNET3D architecture
| DeepUNET3D | Orthogonal | StackGen-Net | |||||
|---|---|---|---|---|---|---|---|
| Axial | Sagittal | Coronal | Axial (E-A) | (E-A) | (E-MV) | ||
| Dice (F1-P) | 0.74 [SD, 0.06] | 0.73 [SD, 0.08] | 0.72 [SD, 0.02] | 0.73 [SD, 0.07] | 0.75 [SD, 0.08] | 0.75 [SD, 0.08] | 0.76 [SD, 0.07] |
| Precision-P | 0.84 [SD, 0.08] | 0.81 [SD, 0.07] | 0.83 [SD, 0.08] | 0.84 [SD, 0.08] | 0.87 [SD, 0.06] | 0.87 [SD, 0.06] | 0.73 [SD, 0.11] |
| Recall-P | 0.66 [SD, 0.08] | 0.67 [SD, 0.10] | 0.64 [SD, 0.12] | 0.78 [SD, 0.09] | 0.66 [SD, 0.10] | 0.67 [SD, 0.10] | 0.79 [SD, 0.1] |
| Precision-L | 0.81 [SD, 0.10] | 0.79 [SD, 0.09] | 0.85 [SD, 0.11] | 0.84 [SD, 0.09] | 0.88 [SD, 0.09] | 0.87 [SD, 0.09] | 0.75 [SD, 0.11] |
| Recall-L | 0.80 [SD, 0.15] | 0.80 [SD, 0.10] | 0.78 [SD, 0.11] | 0.77 [SD, 0.14] | 0.80 [SD, 0.13] | 0.81 [SD, 0.13] | 0.87 [SD, 0.08] |
| F1-L | 0.80 [SD, 0.11] | 0.79 [SD, 0.07] | 0.80 [SD, 0.08] | 0.80 [SD, 0.09] | 0.83 [SD, 0.08] | 0.83 [SD, 0.08] | 0.80 [SD, 0.09] |
| |VD|(%) | 21.2 [SD, 10.5] | 16.9 [SD, 10.8] | 23.5 [SD, 13.0] | 22.7 [SD, 11.2] | 24.3 [SD, 11.3] | 22.3 [SD, 10.9] | 12.3 [SD, 12.7] |
Note:—|VD| indicates absolute volume difference; P, pixel; L, lesion.
↵a Mean [SD] on test cohort 1.