Table 3.

The area under the ROC curve (AUC) with 95% confidence intervals, sensitivity, and specificity with standard deviation for each selected feature (univariate) and whole feature set (multivariable analysis). The “*” sign indicates significant predictive variables by univariate ROC curve analysis at the level of 0.05

Method	Selected variables	AUC (95% CI)	Sensitivity	Specificity	Overall AUC (95% CI), sensitivity (SD), specificity (SD)	DeLong’s test for comparison of two ROC curves
SCAD-penalized SVM	First order-10 percentile	0.55 (0.40–0.71)	0.82	0.30	0.784 (0.64–0.92), 0.591 (0.09), 0.809 (0.10),	Z = 1.189 (p value = 0.264)
	First order-skewness	0.59 (0.50–0.68)*	0.73	0.53
	GLCM-autocorrelation	0.56 (0.40–0.71)	0.64	0.53
	GLCM-cluster shade	0.55 (0.40–0.70)	0.77	0.43
	GLCM-inverse variance	0.60 (0.52–0.75) *	0.68	0.57
	GLSZM-gray-level non-uniformity normalized	0.62 (0.53–0.76) *	0.69	0.62
	GLDM-large dependence high gray-level emphasis	0.59 (0.50–0.73) *	0.55	0.67
	GLRLM-run percentage	0.57 (0.42–0.72)	0.46	0.76
	GLSZM-gray-level non-uniformity	0.60 (0.50–0.74) *	0.68	0.53
	GLRLM-long-run high gray-level emphasis	0.55 (0.39–0.70)	0.55	0.67
RP algorithm	GLCM-sum average	0.56 (0.40–0.71)	0.91	0.24	0.654 (0.50–0.82), 0.727 (0.11), 0.523 (0.09),
	NGTDM-contrast	0.56 (0.40–0.71)	0.86	0.34
	GLSZM-gray-level variance	0.54 (0.38–0.69)	0.64	0.57
	GLSZM-high-gray-level zone emphasis	0.54 (0.38–0.69)	0.59	0.57
	First order-10 percentile	0.55 (0.40–0.71)	0.82	0.30
	First order-skewness	0.59 (0.50–0.68) *	0.73	0.53
	GLSZM-low gray-level zone emphasis	0.55 (0.39–0.70)	0.32	0.76