Table 6.
Comparison of the observers’ LLF according to nodule location without versus with the DLD system
| Session 1 | Session 2 | p-value | |
|---|---|---|---|
| Danger zone nodules | |||
| Thoracic radiologists | 81.7 (178/218) | 81.7 (178/218) | > 0.999 |
| Non-thoracic radiologists | 56.4 (123/218) | 61.9 (135/218) | 0.735 |
| Radiology residents | 56.7 (247/436) | 65.6 (286/436) | 0.021* |
| Non-danger zone nodules | |||
| Thoracic radiologists | 84.6 (181/214) | 86.9 (186/214) | > 0.999 |
| Nonthoracic radiologists | 77.1 (165/214) | 79.9 (171/214) | > 0.999 |
| Radiology residents | 75.5 (323/428) | 74.8 (210/428) | 0.462 |
DLD Deep learning-based nodule detection, LLF Lesion localization fraction
*p < 0.05 was regarded as statistically significant. For group-averaged comparison, corrected p-values are presented (multiplied by 3). Denominator for the LLF calculation corresponds to the total number of true lesions multiplied by the number of readers in the respective group