TABLE 2.
Diagnostic performance of the initial detection by Mask R-CNN, and after classification by ResNet50
| Dataset | Model | TP | FN | Sensitivity | *TN | *Specificity | Total FP | FP from Benign | FP from Vessels | FP from Parenchyma | PPV | FDR |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Dataset-1 | Mask R-CNN | 101 | 2 | 98% | 5 | 9% | 179 | 48 | 33 | 98 | 36% | 64% |
| ResNet50 | 99 | 4 | 96% | 37 | 70% | 32 | 16 | 7 | 9 | 76% | 24% | |
| Dataset-2 | Mask R-CNN | 49 | 4 | 92% | 4 | 13% | 121 | 27 | 23 | 71 | 29% | 71% |
| ResNet50 | 43 | 10 | 81% | 25 | 81% | 28 | 6 | 8 | 14 | 61% | 39% |
*
Specificity is referring to the diagnosis of histologically confirmed benign lesions.TP: True Positive; FN: False Negative;
*
TN: True Negative of confirmed benign lesions; FP: False Positive; Sensitivity = TP/(TP+FN); Specificity = TN/(TN+FP from Begin); PPV: Positive Predictive Value = TP/(TP+Total FP), FDR: False Detection Rate = FP/ (FP+TP).