Table 3. Diagnostic performance of DeepDiscern lesion detection DNN.
Lesion type | Precision rate | Recall rate | F1 score |
Stenosis | 0.769 | 0.901 | 0.829 |
Total occlusion | 0.757 | 0.871 | 0.810 |
Calcification | 0.751 | 0.862 | 0.802 |
Thrombus | 0.742 | 0.925 | 0.823 |
Dissection | 0.790 | 0.926 | 0.854 |
DeepDiscern achieved an average recall of 89.7% for the five lesion types, namely, stenosis, total occlusion, calcification, thrombus and dissection (λd= 0.5). |