Research Article
Deep Learning Models to Predict Fatal Pneumonia Using Chest X-Ray Images
Table 6
Performance measures of the deep learning models and physicians on the external validation test dataset.
| | | AutoML | NNC | Respiratory physician 2 | Respiratory physician 3 | Resident 1 | Resident 2 |
| | Sensitivity (95% CI) | 68.0 (53.3–80.5) | 38.0 (24.7–52.8) | 44.0 (30.0–58.7) | 22.0 (11.5–36.0) | 70.0 (55.4–82.1) | 74.0 (59.7–85.4) |
| | Specificity (95% CI) | 86.0 (73.3–94.2) | 92.0 (80.8–97.8) | 94.0 (83.5–98.7) | 98.0 (89.4–99.9) | 72.0 (57.5–83.8) | 70.0 (55.4–82.1) |
| | PPV (95% CI) | 82.9 (67.9–92.8) | 82.6 (61.2–95.0) | 88.0 (68.8–97.5) | 91.7 (61.5–99.8) | 71.4 (56.7–83.4) | 71.2 (56.9–82.9) |
| | NPV (95% CI) | 72.9 (59.7–83.6) | 59.7 (47.9–70.8) | 62.7 (50.7–73.6) | 55.7 (44.7–66.3) | 70.6 (56.2–82.5) | 72.9 (62.1–80.5) |
| | Accuracy (95% CI) | 77.0 (67.5–84.8) | 65.0 (54.8–74.3) | 69.0 (59.0–77.9) | 60.0 (49.7–69.7) | 71.0 (61.1–79.6) | 72.0 (62.1–80.5) |
| | F1 score (%) | 74.7 | 52.1 | 58.7 | 35.5 | 70.7 | 72.6 |
|
|
CI, confidence interval.
|
