Table 2.

The performance of CT radiomics models in training, cross-validation and testing cohorts

Models	Dataset	AUC (95% CI)	ACC (95% CI)	Specificity (95% CI)	Sensitivity (95% CI)
Clinical model	Training	0.942 (0.916 to 0.969)	0.844 (0.794 to 0.883)	0.87 (0.762 to 0.935)	0.835 (0.776 to 0.882)
	Validation	0.881 (0.835 to 0.927)	0.793 (0.739 to 0.838)	0.826 (0.712 to 0.903)	0.782 (0.718 to 0.835)
	Testing	0.739 (0.58 to 0.898)a	0.647 (0.5 to 0.772)	0.769 (0.46 to 0.938)	0.605 (0.435 to 0.755)
Radiological model	Training	0.922 (0.89 to 0.955)	0.804 (0.751 to 0.848)	0.957 (0.87 to 0.989)	0.752 (0.687 to 0.809)
	Validation	0.869 (0.82 to 0.918)	0.775 (0.72 to 0.822)	0.899 (0.796 to 0.955)	0.733 (0.666 to 0.791)
	Testing	0.818 (0.698 to 0.938)a	0.588 (0.442 to 0.721)	1 (0.717 to 1)	0.447 (0.29 to 0.615)
Radiomic model	Training	0.962 (0.939 to 0.986)	0.909 (0.867 to 0.939)	0.884 (0.779 to 0.945)	0.917 (0.869 to 0.95)
	Validation	0.828 (0.767 to 0.889)	0.825 (0.774 to 0.867)	0.797 (0.68 to 0.881)	0.835 (0.776 to 0.882)
	Testing	0.765 (0.585 to 0.946)	0.667 (0.52 to 0.789)	0.692 (0.389 to 0.896)	0.658 (0.486 to 0.799)
Quantifying model	Training	0.899 (0.863 to 0.935)	0.815 (0.762 to 0.858)	0.812 (0.696 to 0.892)	0.816 (0.754 to 0.865)
	Validation	0.803 (0.742 to 0.863)	0.778 (0.724 to 0.825)	0.725 (0.602 to 0.822)	0.796 (0.733 to 0.848)
	Testing	0.607 (0.414 to 0.8)a	0.608 (0.461 to 0.738)	0.615 (0.323 to 0.849)	0.605 (0.435 to 0.755)
Integrated model	Training	0.984 (0.971 to 0.997)	0.956 (0.923 to 0.976)	0.899 (0.796 to 0.955)	0.976 (0.941 to 0.991)
	Validation	0.893 (0.841 to 0.946)	0.88 (0.834 to 0.915)	0.754 (0.633 to 0.846)	0.922 (0.875 to 0.954)
	Testing	0.925 (0.856 to 0.994)	0.843 (0.709 to 0.925)	0.923 (0.621 to 0.996)	0.816 (0.651 to 0.917)

^aDeLong test showed significant different (p < 0.05) between the model with integrated model on the testing cohort. CI confidence interval