Table 2.
The accuracy, sensitivity, and specificity of the CNN models. Ensemble model consists of two models that produced the best outputs together, which in this case was the EfficientNet B2 for AP and the EfficientNet B4 for lateral radiographs. Data in parentheses are the 95% confidence interval. CNN, convolutional neural network; AP, anteroposterior
| Model | Views | Accuracy | Sensitivity | Specificity | AUC | 95% CI |
|---|---|---|---|---|---|---|
| EfficientNet B2 | AP | 98.7 (95.3–99.6) | 98.7 (92.8–99.8) | 98.7 (92.8–99.8) | 0.995 | 0.971–0.999 |
| Lateral | 93.3 (88.2–96.3) | 92.0 (83.6–96.3) | 94.7 (87.1–97.9) | 0.976 | 0.940–0.990 | |
| EfficientNet B3 | AP | 96.7 (92.4–98.6) | 93.3 (85.3–97.1) | 100 (95.1–100) | 0.993 | 0.980–0.998 |
| Lateral | 93.3 (88.2–96.3) | 94.7 (87.1–97.9) | 92.0 (83.6–96.3) | 0.982 | 0.958–0.992 | |
| EfficientNet B4 | AP | 96.7 (92.4–98.6) | 97.3 (90.8–99.3) | 96.0 (88.9–98.6) | 0.989 | 0.961–0.997 |
| Lateral | 98.7 (95.3–99.6) | 98.7 (92.8–99.8) | 98.7 (92.8–99.8) | 0.993 | 0.955–0.999 | |
| EfficientNet B5 | AP | 98.7 (95.3–99.6) | 98.7 (92.8–99.8) | 98.7 (92.8–99.8) | 0.995 | 0.968–0.999 |
| Lateral | 96.0 (91.5–98.2) | 97.3 (90.8–99.3) | 94.7 (87.1–97.9) | 0.987 | 0.954–0.997 | |
| Ensemble | AP + Lateral | 99.3 (96.3–99.9) | 98.7 (92.8–99.8) | 100 (95.1–100) | 0.993 | 0.949–0.999 |