TABLE 3.

Performance of the “LASSO model” and “mRMR-LASSO model.”

Datasets	Method	Feature count	AUC (95% CI)	ACC	SEN	SPE	p-Value
Training dataset	LASSO	7	0.693 (0.638, 0.747)	0.717	0.811	0.518	0.003
	mRMR-LASSO	7	0.767 (0.718, 0.816)	0.774	0.869	0.574
Temporal validation dataset	LASSO	7	0.767 (0.689, 0.845)	0.735	0.735	0.736	0.092
	mRMR-LASSO	7	0.828 (0.759, 0.897)	0.806	0.928	0.667

p-Values were derived from the DeLong test comparing AUCs between radiomics signatures built by two feature selection methods. AUC, area under the receiver operating curve; ACC, accuracy; CI, confidence interval; SEN, sensitivity; SPE, specificity.