Table 5.
Evaluation of the capabilities of auxiliary tasks to improve the binary breast classification performed on the DHMItest and DINB datasets, while training with fewer cases annotated with biradsbin labels.
| Dataset | ||
|---|---|---|
| D HMItest | D INB | |
| Reference | 80.46 ± 0.79 | 78.13 ± 0.83 |
| MTL breast−25% | 78.53 ± 0.22 | 76.01 ± 1.02 |
| MTL breast−50% | 78.57 ± 0.19 | 74.07 ± 1.11 |
| MTL breast−75% | 76.20 ± 0.39 | 71.89 ± 0.87 |
| MTL all−25% | 77.85 ± 0.50 | 77.01 ± 0.98 |
| MTL all−50% | 76.37 ± 0.49 | 72.32 ± 0.78 |
| MTL all−75% | 74.18 ± 0.37 | 69.97 ± 0.92 |
| MixMatch | 74.85 ± 0.44 | 75.10 ± 0.86 |
| NL 50% | 73.82 ± 0.56 | 72.88 ± 0.77 |
| NL 100% | 72.12 ± 0.67 | 74.55 ± 1.21 |
AUC score is reported.