Table 2. Nodule-Detection Performance of Bone-Suppressed DCNN Model in the Internal and External Validation Tests.
| Variables | Original model | DLBS model | P value |
|---|---|---|---|
| Institute 1 (n = 100) | |||
| Sensitivity, % (No./total No.) | 86.7 (52/60) | 96.7 (58/60) | .008 |
| 95% CI | 78.1-95.3 | 92.1-100 | |
| FPPI | 0.06 (6/100) | 0.05 (5/100) | .71 |
| 95% CI | 0.02-0.15 | 0.02-0.12 | |
| Institute 2 (n = 246) | |||
| Sensitivity, % (No./total No.) | 79.8 (95/119) | 91.5 (109/119) | <.001 |
| 95% CI | 72.7–87.1 | 87.6-97.2 | |
| FPPI | 0.09 (23/246) | 0.07 (17/246) | <.001 |
| 95% CI | 0.06-0.14 | 0.04-0.11 | |
| Institute 3 (n = 205) | |||
| Sensitivity, % (No./total No.) | 80.4 (74/92) | 92.4 (85/92) | <.001 |
| 95% CI | 74.1-90.3 | 87.0-97.8 | |
| FPPI | 0.16 (32/205) | 0.09 (19/205) | <.001 |
| 95% CI | 0.101-0.215 | 0.05-0.15 |
Abbreviations: DCNN, deep convolutional neural network; DLBS, deep learning–based synthetic bone-suppressed; FPPI, false-positive findings per image.