Table 1 Effectiveness of our proposed method on the dice coefficient, precision, for methods including the primary network, two-stage network, and augmentation(Aug) + oversample strategies
From: A deep semantic segmentation correction network for multi-model tiny lesion areas detection
Method | Task | Segmentation | Detection | Classification |
|---|---|---|---|---|
Primary network | a | 10.23 | 9.25 | – |
b | 32.92 | 16.12 | – | |
Primary network + Aug | a | 15.83 | 12.96 | – |
b | 29.56 | 15.80 | – | |
Primary network + Oversample | a | 34.66 | 24.07 | – |
b | 54.39 | 58.06 | – | |
Primary network + Aug + Oversample | a | 32.48 | 52.18 | - |
b | 67.50 | 80.64 | – | |
Two-stage network | b | 26.53 | 28.23 | 66.97 |
Two-stage network + Aug | b | 30.17 | 58.40 | 75.21 |
Two-stage network + Oversample | b | 67.82 | 76.47 | 87.45 |
Proposed method | b | 74.21 | 91.76 | 92.89 |