AUC results for each of the three centres individually and for the whole dataset overall
| Site | Nodules (n) | Cancer (%) | LCP-CNN (95% CI) | Brock (95% CI) | Maximal axial diameter (95% CI) |
| Leeds | 438 | 20.3 | 88.1 (84.3 to 91.5) | 83.4 (78.9 to 87.5) | 77.6 (72.5 to 82.4) |
| Nottingham | 423 | 22.0 | 89.0 (85.3 to 92.3) | 87.3 (82.8 to 91.2) | 82.3 (77.0 to 87.4) |
| Oxford | 536 | 9.7 | 91.9 (88.5 to 94.9) | 91.3 (87.3 to 94.6) | 91.0 (86.8 to 94.4) |
| All | 1397 | 16.8 | 89.6 (87.6 to 91.5)* | 86.8 (84.3 to 89.1)* | 83.1 (80.3 to 85.8) |
‘Diameter’ simply means stratifying the nodules according to the maximal diameter on an axial slice and using that to create a receiver operating curve.
*P=0.0044 for LCP-CNN versus Brock.
AUC, area under the curve; LCP-CNN, lung cancer prediction convolutional neural network.