Table 3 Performance of algorithms in the detection of invasive and in situ breast carcinoma.
Study | Slides/case | Invasive carcinoma accuracy | In situ carcinoma accuracy |
|---|---|---|---|
Cruz-Roa et al., 201719 | 195 cancer cases (TGGA), 21 normal | F1 = 75.9% | Not available |
PPV = 71.6% | |||
NPV = 96.8% | |||
Han et al., 201720 | 21 cases (BreaKHis) | Accuracy = 0.93 | Not available |
Bejnordi et al., 201817 | 330 cases (928 WSIs) | AUC = 0.96 | Not available |
Mercan et al., 201916 | 240 cases | Accuracy = 0.94 | Accuracy = 0.70 |
Sens = 70% | Sens = 79% | ||
Spec = 95% | Spec = 41% | ||
Sheikh et al., 202028 | 92 WSIs (ICIAR2018) | Accuracy = 0.68 | Accuracy = 0.64 |
Max Sens = 96% | Max Sens = 83% | ||
Max Spec = 85% | Max Spec = 93% | ||
Polónia et al., 202127 | 152 ROIs (10 WSIs from 8 cases) | Accuracy = 0.92 | Accuracy = 0.88 |
Current study | 436 cases (841 WSIs) | AUC = 0.99 | AUC = 0.98 |
Sens = 95.5% | Sens = 93.2% | ||
Spec = 93% | Spec = 93.8% |
