Table 2.
Evaluation of segmentation results obtained using existing methods (DeepLab v3+ and U-net) and CSC (Cropping-Segmentation-Calibration) with respect to the mIoU and mean Dice coefficient (mDice).
| Method | mIoU | mDice | ||
|---|---|---|---|---|
| Original Image | Cropped Image | Original Image | Cropped Image | |
| DeepLab v3+ | 0.0224 ± 0.0085 | 0.0382 ± 0.0140 | ||
| U-net | 0.1519 ± 0.0596 | 0.2238 ± 0.0777 | ||
| CSC (Ours) | 0.5543 ± 0.0081 | 0.5598 ± 0.0067 | 0.6891 ± 0.0104 | 0.6950 ± 0.0074 |
The values are the mean ± standard deviation of the three datasets for cross-validation. CSC yielded the highest values. Moreover, the cropped images yielded slightly higher values than the original images in CSC.