Table 2 Sensitivity, specificity, positive predictive value, and negative predictive value in the binary-class classification.
Test dataset | Sensitivity | Specificity | PPV | NPV |
|---|---|---|---|---|
Binary Classification at High Sensitivity Threshold | ||||
RD/1,282 images | 67.5 (58.5–75.6) | 77.0 (74.4–79.3) | 23.6 (20.8–26.6) | 95.7 (94.6–96.7) |
RDadequate/787 images | 74.1 (64.7–82.4) | 73.1 (69.9–76.4) | 25.0 (21.8–28.3) | 95.9 (94.5–97.2) |
RDinadequte/495 images | 52.6 (36.8–68.4) | 82.7 (79.4–86.2) | 20.2 (14.3–26.3) | 95.5 (93.9–97.0) |
SNU/2,201 images | 90.1 (85.1–94.0) | 91.7 (90.4–92.8) | 50.6 (46.9–54.5) | 99.0 (98.5–99.4) |
SNU subset/240 images | 85.0 (75.0–95.0) | 94.0 (90.5–97.0) | 74.4 (64.0–85.7) | 96.9 (94.9–99.0) |
Edinburgh/1,300 images | 97.7 (96.3–99.0) | 52.0 (48.5–55.6) | 54.5 (52.7–56.4) | 97.5 (95.9–98.8) |
Binary Classification at High Specificity Threshold | ||||
RD/1,282 images | 44.7 (36.6–53.7) | 91.8 (90.2–93.4) | 36.7 (30.7–43.5) | 94.0 (93.2–94.9) |
RDadequate/787 images | 51.8 (41.2–62.4) | 90.6 (88.3–92.7) | 40.0 (32.7–47.8) | 94.0 (92.7–95.2) |
RDinadequte/495 images | 29.0 (15.8–42.1) | 93.7 (91.5–95.8) | 27.3 (15.8–40.5) | 94.0 (93.0–95.2) |
SNU/2,201 images | 80.8 (74.7–86.8) | 95.9 (95.0–96.8) | 65.5 (60.1–70.6) | 98.1 (97.6–98.7) |
SNU subset/240 images | 77.5 (62.5–90.0) | 95.0 (92.0–97.5) | 76.2 (64.6–86.9) | 95.5 (92.8–97.9) |
Edinburgh/1,300 images | 90.6 (87.9–93.1) | 77.3 (74.6–80.2) | 70.1 (67.7–72.9) | 93.4 (91.6–95.0) |
