Table 2.
Comparison of the diagnostic performance of each radiologist with and without the DL model. AUC was calculated using ROC (receiver operating characteristic) analysis.
| (Radiology experience) | DL | AUC | p | Accuracy (%) | p | Sensitivity (%) | p | Specificity (%) | p |
|---|---|---|---|---|---|---|---|---|---|
| R1 (1 year) | (−) | 0.62 | 0.08 | 61.9 | 0.05 | 58.8 | 0.72 | 64.1 | 0.002* |
| (+) | 0.66 | 68.1 | 55.9 | 77.2 | |||||
| R2 (6 years) | (−) | 0.68 | 0.49 | 69.4 | 1.00 | 57.3 | 1.00 | 78.3 | 1.00 |
| (+) | 0.67 | 68.8 | 55.9 | 78.3 | |||||
| R3 (12 years) | (−) | 0.65 | 0.31 | 67.5 | 0.57 | 48.5 | 0.14 | 81.5 | 0.55 |
| (+) | 0.69 | 70.0 | 58.8 | 78.3 | |||||
| DL | 0.69 | 71.9 | 53.0 | 85.0 |
*Significantly different from McNemar’s test with Bonferroni correction.
Abbreviations: DL, deep learning; AUC, area under the receiver operating characteristic curve.