Table 1 Diagnostic performance of DCNN models (learning rate = 1e−4).
From: A convolutional deep learning model for improving mammographic breast-microcalcification diagnosis
DCNN models | Cut-off point | AUC (95% CI) | Sensitivity (95% CI) | Specificity (95% CI) | Accuracy (95% CI) | PPV (95% CI) | NPV (95% CI) |
|---|---|---|---|---|---|---|---|
ResNet-101 | > 0.056 | 0.837 (0.784–0.890) | 64.95 (55.45–74.45) | 91.41 (87.11–95.71) | 81.54 (76.82–86.26) | 81.82 (73.21–90.43) | 81.42 (75.78–87.06) |
Xception | > 0.204 | 0.817 (0.760–0.874) | 65.98 (56.55–75.41) | 88.34 (83.41–93.27) | 80.00 (75.14–84.86) | 77.11 (68.07–86.15) | 81.36 (75.62–87.10) |
Inception-v3 | > 0.168 | 0.792 (0.731–0.853) | 77.32 (68.99–85.65) | 77.91 (71.54–84.28) | 77.69 (72.63–82.75) | 67.57 (58.86–76.28) | 85.23 (79.53–90.93) |
Inception- ResNet-v2 | > 0.276 | 0.838 (0.787–0.889) | 75.26 (66.67–83.85) | 80.98 (74.96–87.00) | 78.85 (73.89–83.81) | 70.19 (61.40–78.98) | 84.62 (78.96–90.28) |
DenseNet-201 | > 0.017 | 0.832 (0.782–0.881) | 82.47 (74.90–90.04) | 69.33 (62.25–76.41) | 74.23 (68.91–79.55) | 61.54 (53.18–69.90) | 86.92 (81.12–92.72) |
Ensemble | > 0.247 | 0.856 (0.806–0.907) | 72.16 (63.24–81.08) | 86.50 (81.25–91.75) | 81.15 (76.40–85.90) | 76.09 (67.37–84.81) | 83.93 (78.38–89.48) |
p values | < .0001 | 0.0011 | < .0001 | 0.0870 | < .0001 | 0.1293 |
