Skip to main content
. 2022 Feb 11;32(7):4868–4878. doi: 10.1007/s00330-022-08558-0

Table 3.

Inter-rater agreement for both image-wise and for full examination for both readers and the dCNN compared to ground truth

Accuracy [%] κ (95% CI) Agreement
Single images dCNN/actuals 79.7% 0.57 (0.50–0.64) Moderate
Reader 1/actuals 80.5% 0.61 (0.53–0.68) Substantial
Reader 2/actuals 72.7% 0.44 (0.36–0.51) Moderate
Reader 1/Reader 2 71.9% 0.43 (0.28–0.58) Moderate
Full imageset dCNN/actuals 90.9% 0.82 (0.69–0.95) Almost perfect
Reader 1/actuals 81.8% 0.63 (0.47–0.81) Substantial
Reader 2/actuals 90.9% 0.82 (0.69–0.95) Almost perfect
Reader 1/Reader 2 72.7% 0.46 (0.09–0.82) Moderate