TABLE 1.
Performance of deep learning (DL)‐based auto‐segmentation models (compared with the reference contours)
DL‐based models for different targets | DSC | MSD (mm) | HD (mm) | HD 95 (mm) |
---|---|---|---|---|
dCTV1 | 0.88 ± 0.03 | 1.32 ± 0.48 | 21.60 ± 7.50 | 4.86 ± 0.56 |
dCTV2 | 0.70 ± 0.09 | 2.42 ± 1.62 | 22.44 ± 8.49 | 6.47 ± 1.92 |
pCTV1 | 0.86 ± 0.03 | 1.15 ± 0.38 | 20.78 ± 6.22 | 4.11 ± 0.65 |
Abbreviations: dCTV, clinical tumor volume for definitive radiotherapy; DSC, dice similarity coefficient; HD, Hausdorff distance; HD 95, Hausdorff distance 95%; MSD, mean surface distance; pCTV, clinical tumor volume for postoperative radiotherapy.