Table 4. The AUC of each model validated in different pathological stages. Because sample size (n=9) of stage II in validation set is too small to evaluate, we combine stage I and II for analysis.

Model	AUC in stage I and II (n=63)	AUC in stage III (n=76)	AUC in stage IV (n=61)
MClinical	0.630	0.800	0.617
MRadiomics	0.755	0.742	0.736
MRadiomics+Clinical	0.713	0.838	0.718
MMCNNs	0.767	0.815	0.848
MMCNNs+Clinical	0.800	0.860	0.847
MRadiomics+MCNNs	0.771	0.818	0.839
MRadiomics+MCNNs+Clinical	0.803	0.861	0.854

MCCNs, multi-level residual CNNs; CNN, convolutional neural network; AUC, area under the receiver operating characteristic curve.