Table 4 Performances of the deep learning models for slice-level diagnosis.
From: A novel deep learning model for a computed tomography diagnosis of coronary plaque erosion
AUC (95% CI) | Sensitivity (%) (95% CI) | Specificity (%) (95% CI) | PPV (95% CI) | NPV (95% CI) | FPR (%) (95% CI) | FNR (%) (95% CI) | |
|---|---|---|---|---|---|---|---|
Five-fold cross-validation | |||||||
MD-CTA model | 0.891 (0.887–0.895) | 82.9 (81.7–84.0) | 80.0 (79.5–80.5) | 38.5 (37.9–39.1) | 96.9 (96.7–97.1) | 20.0 (19.5–20.5) | 17.1 (16.0–18.3) |
CNN model | 0.729 (0.722–0.737) | 66.2 (64.8–67.6) | 66.8 (66.3–67.4) | 23.2 (22.7–23.7) | 92.9 (92.6–93.2) | 33.2 (32.6–33.7) | 33.8 (32.4–35.2) |
Test set validation | |||||||
MD-CTA model | 0.897 (0.890–0.904) | 82.2 (79.8–84.3) | 80.1 (79.1–81.0) | 39.3 (38.0–40.5) | 96.6 (96.2–97.0) | 19.9 (19.0–80.9) | 17.8 (15.7–20.2) |
CNN model | 0.757 (0.744–0.770) | 68.9 (66.2–71.6) | 67.3 (66.3–68.4) | 24.9 (23.9–25.8) | 93.3 (92.7–93.8) | 32.7 (31.6–33.7) | 31.1 (28.4–33.8) |
