Virtual biopsy using CT radiomics for evaluation of disagreement in pathology between endoscopic biopsy and postoperative specimens in patients with gastric cancer: a dual-energy CT generalizability study

Insights into Imaging

Table 2 Performance of five radiomics signature on the training and test sets

Cohort	AUC [95% CI]	F1 score	Sensitivity	Specificity	NPV	PPV
LR
Training set	0.92 [0.876–0.973]	0.80	0.77	0.91	0.86	0.85
Test set	0.77 [0.584–0.961]	0.61	0.64	0.72	0.76	0.58
SVM
Training set	0.91 [0.841–0.973]	0.83	0.81	0.91	0.85	0.89
Test set	0.83 [0.671–0.986]	0.67	0.64	0.83	0.7	0.79
DT
Training set	0.85 [0.774–0.916]	0.72	0.81	0.72	0.86	0.65
Test set	0.71 [0.499–0.914]	0.52	0.64	0.5	0.69	0.44
SGD
Training set	0.77 [0.679–0.857]	0.63	0.70	0.67	0.78	0.57
Test set	0.74 [0.536–0.950]	0.67	0.73	0.72	0.81	0.62
KNN
Training set	0.74 [0.655–0.833]	0.41	0.28	0.94	0.68	0.75
Test set	0.63 [0.435–0.833]	0.27	0.19	0.89	0.64	0.5

AUC area under the curve; CI confidence interval; NPV negative predictive value; PPV positive predictive value; LR logistic regression; SVM support vector machine; DT decision tree; SGD stochastic gradient descent; KNN k-nearest neighbors