Table III.

LASSO-SVM model prediction results in terms of area under the curve (AUC), accuracy (Acc), false positive rate (FPR) and true positive rate (TPR). Performance on the training set summarizes predictions of the 10×10-fold CV loops for the model based on no-harmonized features (“without harmonization”) and the one based on harmonized features (“with harmonization”). External-validation results are instead subdivided according to the three scenarios performed (A, B and C).

	Cross-validation		External validation
	no-harmonized features	harmonized features	scenario A	scenario B	scenario C
AUC (95% CI)	0.90 (0.85–0.96)	0.89 0.84–0.95)	0.86 (0.78–0.95)	0.86 (0.77–0.95)	0.86 (0.77–0.95)
Acc [%]	78.7	80.5	79.1	81.9	79.1
FPR [%]	21.9	20.0	35.5	25.9	32.3
TPR [%]	79.3	81.0	90.0	87.8	87.8