Research Article
Cobb Angle Measurement of Spine from X-Ray Images Using Convolutional Neural Network
Table 2
Dice similarity coefficient (DSC) from 5-fold cross-validation of U-Net, Residual U-Net, and Dense U-Net.
| k-fold | Dice similarity coefficient (DSC) | U-Net | Residual U-Net | Dense U-Net |
| k = 1 | 0.940 ± 0.036 | 0.952 ± 0.023 | 0.947 ± 0.028 | k = 2 | 0.942 ± 0.032 | 0.951 ± 0.029 | 0.947 ± 0.029 | k = 3 | 0.942 ± 0.033 | 0.952 ± 0.025 | 0.949 ± 0.028 | k = 4 | 0.941 ± 0.034 | 0.949 ± 0.030 | 0.947 ± 0.026 | k = 5 | 0.942 ± 0.035 | 0.952 ± 0.028 | 0.947 ± 0.030 | Average ± std. | 0.941 ± 0.034 | 0.951 ± 0.027 | 0.948 ± 0.028 | Parameter size | 1.21 million | 1.19 million | 1.20 million | Training time | 0.34 hour | 0.77 hour | 2.33 hour | Testing time (each image) | 0.03 second | 0.05 second | 0.07 second |
|
|