Table 2.
JAFROC figure of merit | FPPI | ||||||
---|---|---|---|---|---|---|---|
Session 1 | p-value (vs. DLD) | Session 2 | p-value (vs. session 1) | Session 1 | Session 2 | p-value | |
DLD system | 0.836 | 0.20 | |||||
Thoracic radiologists | 0.895 | 0.006* | 0.932 | < 0.001* | 0.17 (104/600) | 0.12 (73/600) | < 0.001* |
Non-thoracic radiologists | 0.850 | 0.777 | 0.860 | > 0.999 | 0.12 (72/600) | 0.08 (47/600) | < 0.001* |
Radiology residents | 0.806 | 0.204 | 0.862 | 0.009* | 0.19 (232/1200) | 0.12 (140/1200) | < 0.001* |
DLD Deep learning-based nodule detection, FPPI False-positive findings per image, JAFROC Jackknife alternative free-response receiver operating characteristic
*p < 0.05 was regarded as statistically significant. For group-averaged comparison, corrected p-values are presented (multiplied by 3)