Skip to main content
. 2021 Mar 11;4:48. doi: 10.1038/s41746-021-00417-4

Table 2.

Performance of the hierarchical deep learning system in different confidence regions on validation datasets 1 and 2.

Number of images % (95% CI)
Datasets Regions True False True False Sensitivity Specificity
Positive Negative Negative Positive
Overall region 2770 114 3210 207 0.9605 (0.953-0.968) 0.9394 (0.931-0.947)
validation dataset 1 Reliable region 2019 48 2455 55 0.9768 (0.970-0.983) 0.9781 (0.972-0.984)
Suspicious region 751 66 755 152 0.9192 (0.900-0.938) 0.8324 (0.808-0.857)
Overall region 592 27 1265 80 0.9564 (0.940-0.973) 0.9405 (0.928-0.953)
validation dataset 2 Reliable region 493 8 929 17 0.9840 (0.973-0.995) 0.9820 (0.974-0.991)
Suspicious region 99 19 336 63 0.8390 (0.772-0.906) 0.8421 (0.806-0.878)