Skip to main content
. 2021 Jul 20;11:679764. doi: 10.3389/fonc.2021.679764

Figure 3.

Figure 3

ROCs of 3D CNN models for the prediction of PFS in the validation group. (A) The ROCs of 3D CNN model trained from scratch, using transfer learning based on nodule or EGFR classification models. The corresponding AUCs were 0.668, 0.701, and 0.744, receptively. (B) The ROCs of only using clinical features, 3D CNN (using transfer learning based on EGFR classification model), and the combination of 3D CNN and clinical features, and the corresponding AUCs were 0.624, 0.744, and 0.771.