Table 3 Classification performances of accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC) by the 3D distance-aware networks with different configurations of input volumes of the original CBCT image, segmentation mask, and SDM by Dense121 U-Net, Attention U-Net, simple U-Net, and SegNet. SDM: signed distance map
Segmentation Model | CBCT image | Segmentation mask | SDM | Accuracy | Sensitivity | Specificity | AUC |
|---|---|---|---|---|---|---|---|
|  | ✔ |  |  | 0.32 | 0.10 | 0.73 | 0.34 |
SegNet | ✔ | ✔ |  | 0.48 | 0.25 | 0.91 | 0.69 |
simple U-Net | ✔ | ✔ |  | 0.77 | 0.80 | 0.73 | 0.83 |
Attention U-Net | ✔ | ✔ |  | 0.52 | 0.40 | 0.73 | 0.62 |
Dense U-Net | ✔ | ✔ |  | 0.71 | 1.00 | 0.18 | 0.69 |
SegNet |  |  | ✔ | 0.77 | 0.65 | 1.00 | 0.91 |
simple U-Net |  |  | ✔ | 0.84 | 0.80 | 0.91 | 0.96 |
Attention U-Net |  |  | ✔ | 0.71 | 0.85 | 0.45 | 0.85 |
Dense U-Net |  |  | ✔ | 0.84 | 0.90 | 0.73 | 0.92 |
SegNet | ✔ |  | ✔ | 0.65 | 0.65 | 0.64 | 0.76 |
simple U-Net | ✔ |  | ✔ | 0.68 | 1.00 | 0.09 | 0.92 |
Attention U-Net | ✔ |  | ✔ | 0.90 | 0.90 | 0.91 | 0.98 |
Dense U-Net | ✔ |  | ✔ | 1.00 | 1.00 | 1.00 | 1.00 |