Table 4
Segmentation scores (350.3–353.5°).
| Scores(%)╲ | Model | UNet [10−2] | UNet [10−3] | UNet [10−4] | UNet [10−5] | UNet++ [10−3] | UNet++ [10−4] | UNet++ [10−5] |
|---|---|---|---|---|---|---|---|---|
| Precision at 0.8 | 99.67 | 99.76 | 99.84 | 99.89 | 99.71 | 99.76 | 99.65 | |
| Precision at 0.5 | 97.45 | 98.06 | 98.24 | 97.11 | 97.87 | 97.87 | 95.13 | |
| Precision at 0.2 | 88.10 | 89.75 | 88.79 | 60.42 | 90.12 | 87.59 | 78.58 | |
| Precision at thropt | 96.01 | 95.63 | 96.45 | 96.38 | 96.06 | 96.48 | 94.32 | |
| Recall at 0.8 | 79.15 | 79.45 | 79.10 | 72.96 | 78.34 | 79.71 | 75.14 | |
| Recall at 0.5 | 91.15 | 90.86 | 91.63 | 91.75 | 90.45 | 91.73 | 90.97 | |
| Recall at 0.2 | 96.55 | 96.27 | 96.83 | 98.10 | 96.09 | 96.82 | 97.24 | |
| Recall at thropt | 92.86 | 93.91 | 93.86 | 92.49 | 93.17 | 93.37 | 91.82 | |
| Dice at 0.8 | 88.24 | 88.45 | 88.27 | 84.33 | 87.74 | 88.62 | 85.67 | |
| Dice at 0.5 | 94.19 | 94.33 | 94.82 | 94.35 | 94.02 | 94.70 | 93.00 | |
| Dice at 0.2 | 92.13 | 92.89 | 92.64 | 74.79 | 93.01 | 91.98 | 86.92 | |
| Diceopt | 94.41 | 94.76 | 95.14 | 94.40 | 94.59 | 94.90 | 93.05 | |
| MS at 0.8 | 4.25 | 4.96 | 4.96 | 7.79 | 4.96 | 5.31 | 6.02 | |
| MS at 0.5 | 1.06 | 1.24 | 1.24 | 1.42 | 1.59 | 1.42 | 1.06 | |
| MS at 0.2 | 0.35 | 0.18 | 0.35 | 0.18 | 0.35 | 0.18 | 0.18 | |
| MS at thropt | 0.53 | 0.35 | 0.53 | 1.24 | 0.53 | 0.53 | 0.88 | |
| thropt | 0.41 | 0.35 | 0.38 | 0.48 | 0.37 | 0.42 | 0.47 |
Notes. Segmentation scores evaluated on the dense zone of 350.3–353.5° segmented by the models reported in Fig. 14. Precision, recall, dice index, and MS rate are evaluated at classification threshold values of 0.8, 0.5, 0.2, and the optimal threshold. The latter refers to the threshold value optimizing the dice index and estimated using P-R curves (see Fig. 15a). Blue (red) refers to the best (least performing) model in each row. The bold scores correspond to the absolute best (if in blue) or lowest (if in red) performance per score.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.