Table 1.
Segmentation results of networks under different SSL settings.
| Settings | Train HQA | Train LQA | AIDE | PP | DSC (%) | RAVD (%) | ASSD (mm) | MSSD (mm) |
|---|---|---|---|---|---|---|---|---|
| S01_30_0_F_NP | 30 | 0 | No | No | 70.1 | 42.0 | 16.1 | 176.3 |
| S02_30_0_F_YP | 30 | 0 | No | Yes | 75.6 | 22.0 | 7.68 | 54.8 |
| S03_331_0_F_NP | 331 | 0 | No | No | 87.9 | 10.4 | 4.65 | 65.1 |
| S04_331_0_F_YP | 331 | 0 | No | Yes | 88.5 | 10.8 | 3.64 | 43.8 |
| S05_30_301_F_NP | 30 | 301 | No | No | 78.4 | 19.0 | 7.92 | 95.3 |
| S06_30_301_F_YP | 30 | 301 | No | Yes | 79.7 | 21.1 | 6.10 | 53.4 |
| S07_30_301_A_NP | 30 | 301 | Yes | No | 79.8 | 18.5 | 10.8 | 116.3 |
| S08_30_301_A_YP | 30 | 301 | Yes | Yes | 82.9 | 16.9 | 5.43 | 49.8 |
| S09_30_954_F_NP | 30 | 954 | No | No | 79.5 | 19.7 | 8.43 | 104.2 |
| S10_30_954_F_YP | 30 | 954 | No | Yes | 80.2 | 19.1 | 6.47 | 53.8 |
| S11_30_954_A_NP | 30 | 954 | Yes | No | 86.1 | 10.2 | 5.49 | 75.8 |
| S12_30_954_A_YP | 30 | 954 | Yes | Yes | 86.9 | 10.0 | 4.17 | 44.6 |
| Pseudo-label32 | 30 | 954 | No | Yes | 81.4 | 19.7 | 5.99 | 52.9 |
| Co-teaching27 | 30 | 954 | No | Yes | 82.4 | 15.8 | 5.50 | 49.2 |
HQA and LQA indicate high-quality and low-quality annotations. LQAs are generated by the model trained using the data provided with HQAs. PP refers to post-processing, which is the process of keeping the largest connected components. Context is included in the notation of the experimental setting. For the setting S05_30_301_F_NP, 30 refers to 30 training samples with HQAs, 301 means 301 training samples with LQAs, F indicates training with the conventional fully supervised learning approach, and NP means no post-processing.