Table 1 Summarized results of performance assessments of DL algorithm and radiologists.
From: Development and clinical application of deep learning model for lung nodules screening on CT images
Data set | Rater | Time consumption per study (mins) | FROC score (95% CI) | ROC Sensitivity % (95% CI) | ROC Specificity % (95% CI) | AUC (95% CI) |
|---|---|---|---|---|---|---|
LUNA (n = 888) | DL algorithm | 0.08 | 0.80 [0.78, 0.82] | 82 [73, 94] | 82 [70, 91] | 0.90 [0.88, 0.92] |
Multi-centre validation (n = 582) | DL algorithm | 0.09 | 0.75 [0.73, 0.78] | 73 [63, 86] | 85 [73, 96] | 0.86 [0.84, 0.90] |
Radiologists | 1.71* [0.87, 3.23] | NA§ | 83 [75, 89] | 64 [51, 77] | 0.73 [0.68, 0.78] |
