Evaluation metrics
We evaluate the performances of the models using the following metrics, Accuracy (Acc), Sensitivity (Sens), Specificity (Spec), F1, Area Under the Curve (AUC)
Ranking of singletask approaches
The singletask approaches are trained on the benchmark dataset with 3,641 BUS images. Click on a metric to sort approaches based on that metric.
| Rank | Approaches | Acc | Sens | Spec | F1 | AUC |
|---|---|---|---|---|---|---|
| 1 | VGG16 | 74.5 | 86.7 | 62.6 | 0.77 | 74.7 |
| 2 | MobileNet | 74.0 | 87.4 | 61.3 | 0.77 | 74.4 |
| 3 | Xception | 73.7 | 88.5 | 59.6 | 0.77 | 74.0 |
| 4 | EfficientNetB0 | 73.8 | 86.8 | 61.2 | 0.77 | 74.0 |
| 5 | InceptionV3 | 73.0 | 88.4 | 57.6 | 0.77 | 73.0 |
| 6 | DenseNet121 | 72.7 | 90.1 | 55.7 | 0.77 | 72.9 |
| 7 | Tanaka | 77.8 | 74.6 | 81.2 | 0.76 | 77.9 |
| 8 | ResNet50 | 72.6 | 86.2 | 59.4 | 0.76 | 72.8 |
| 9 | Shia | 74.6 | 75.5 | 74.0 | 0.75 | 74.7 |
| 10 | Xie | 62.2 | 48.6 | 75.8 | 0.55 | 62.2 |
Ranking of multitask approaches
The multitask approaches are trained on BUSI and BUSIS datasets combined with 1,209 BUS images.
| Rank | Approaches | Acc | Sens | Spec | F1 | AUC |
|---|---|---|---|---|---|---|
| 1 | MT-ESTAN (ours) | 90.0 | 90.4 | 89.8 | 0.88 | 90.1 |
| 2 | MobileNet | 87.0 | 81.1 | 91.0 | 0.83 | 86.1 |
| 3 | VGG16 | 87.1 | 81.3 | 90.9 | 0.83 | 86.1 |
| 4 | EfficientNetB0 | 87.5 | 81.0 | 91.2 | 0.83 | 86.1 |
| 5 | Zhang | 87.4 | 81.4 | 91.4 | 0.83 | 86.4 |
| 6 | ResNet50 | 86.1 | 80.9 | 89.2 | 0.81 | 85.0 |
| 7 | DenseNet121 | 85.0 | 79.1 | 88.9 | 0.80 | 84.0 |
| 8 | Shi | 83.9 | 87.3 | 81.7 | 0.80 | 84.5 |
| 9 | Vakanski | 83.6 | 77.4 | 87.8 | 0.78 | 82.6 |