Table 2 Diagnostic performances of CNN and endoscopists
From: Convolutional neural network-based system for endocytoscopic diagnosis of early gastric cancer
|  |  | Per–image | Per–lesion |
|---|---|---|---|
Accuracy, % (95% CI) | CNN | 83.2 (79.8–86.2) | 86.1 (75.9–93.1) |
Endoscopists | 76.8 (74.6–78.8) | 82.4 (76.7–87.2) | |
Sensitivity, % (95% CI) | CNN | 76.4 (71.3–81.0) | 82.1 (66.5–92.5) |
Endoscopists | 73.4 (70.4–76.2) | 79.5 (71.0–86.4) | |
Specificity, % (95% CI) | CNN | 92.3 (88.2–95.4) | 90.9 (75.7–98.1) |
Endoscopists | 81.3 (78.2–84.1) | 85.9 (77.4–92.0) | |
PPV, % (95% CI) | CNN | 93.0 (89.2–95.8) | 91.4 (76.9–98.2) |
Endoscopists | 83.9 (81.2–86.4) | 86.9 (79.0–92.7) | |
NPV, % (95% CI) | CNN | 74.6 (69.2–79.5) | 81.1 (64.8–92.0) |
Endoscopists | 69.6 (66.4–72.8) | 78.0 (69.0–85.4) |