Table 2.

Predictive performance of several deep learning models in the test set

Model	AUC (95%CI)	Accuracy	Sensitivity	Specificity	PPV	NPV
CT_origin	0.544 (0.435–0.653)	0.536	0.507	0.585	0.679	0.407
CT_TL	0.701 (0.595–0.808)	0.688	0.746	0.585	0.757	0.571
PET_origin	0.573 (0.461–0.684)	0.536	0.521	0.561	0.673	0.404
PET_TL	0.645 (0.534–0.756)	0.589	0.549	0.659	0.736	0.458
DS_TL	0.722 (0.622–0.822)	0.661	0.676	0.634	0.762	0.531
TS_TL	0.730 (0.629–0.830)	0.670	0.676	0.659	0.774	0.540

Bold numbers indicate the best results for each evaluation metric

AUC Area under the receiver operating characteristic curve, PPV positive predictive value, NPV Negative predictive value, CT_origin CT model from scratch, CT_TL CT transfer learning, PET_origin PET model from scratch, PET_TL PET transfer learning, DS_TL dual-stream transfer learning, TS_TL three-stream transfer learning