Table 3.

Predictive performance of subjective findings, radiomics signature and radiomics nomogram models for differentiating lung adenocarcinomas and granulomatous lesions in patients with SCSNs

	Training set (n = 150)			External validation set (n = 64)
	Subjective findings model	Radiomics signature	Radiomics nomogram	Subjective findings model	Radiomics signature	Radiomics nomogram
AUC (95% CI)	0.762 (0.686–0.828)	0.834 (0.764–0.889)	0.885 (0.823–0.931)	0.619 (0.489–0.738)	0.798 (0.679–0.888)	0.808 (0.690–0.896)
Sensitivity	0.831 (64/77)	0.766 (59/77)	0.727 (56/77)	0.657 (23/35)	0.714 (25/35)	0.714 (25/35)
Specificity	0.603 (44/73)	0.781 (57/73)	0.904 (66/73)	0.621 (18/29)	0.828 (24/29)	0.828 (24/29)
Accuracy	0.720 (108/150)	0.773 (116/150)	0.813 (122/150)	0.641 (41/64)	0.766 (49/64)	0.766 (49/64)
PPV	0.688 (64/93)	0.787 (59/75)	0.889 (56/63)	0.676 (23/34)	0.833 (25/30)	0.833 (25/30)
NPV	0.772 (44/57)	0.760 (57/75)	0.759 (66/87)	0.600 (18/30)	0.706 (24/34)	0.706 (24/34)

Note. CI Confidence interval; AUC Area under curve; NPV Negative predictive value; PPV Positive predictive value. Numbers in the parentheses were used to calculate percentages. SCSNs Sub-centimeter solid nodules