Table 2.
Diagnostic performance of deep convolutional neural network model for differentiating breast cancer subtypes in the main cohort.
| Experiment | Model | AUC | ACC (%) | SEN (%) | SPC (%) | YI (%) |
|---|---|---|---|---|---|---|
| Luminal A vs. non-luminal A | EfficientNet-B0 | 0.686 | 78.1 | 86.5 | 48.9 | 35.4 |
| DenseNet-121 | 0.717 | 80.1 | 87.8 | 53.3 | 41.1 | |
| VGGNet-19 | 0.664 | 74.6 | 82.1 | 48.9 | 31.0 | |
| Luminal vs. non-luminal | EfficientNet-B2 | 0.601 | 64.2 | 53.7 | 68.4 | 22.1 |
| DenseNet-121 | 0.587 | 61.1 | 64.8 | 59.6 | 24.4 | |
| VGGNet-19 | 0.561 | 64.2 | 48.1 | 70.6 | 18.7 | |
| Triple-negative vs. non-triple-negative | EfficientNet-B2 | 0.577 | 76.9 | 33.3 | 86.3 | 19.6 |
| DenseNet-121 | 0.565 | 58.1 | 60.6 | 57.5 | 18.1 | |
| VGGNet-19 | 0.572 | 50.5 | 69.7 | 46.4 | 16.1 |
AUC, area under the curve; ACC, accuracy; SEN, sensitivity; SPC, specificity; YI, Youden’s index.