Table 3.

Models’ results in terms of AUC, accuracy, precision, and recall.

Harmo CT + Original PET Features (A)
	Linear SVM (A1)
	AUC *	Accuracy	Precision **	Recall **	p ***
Training dataset	0.77 [0.66–0.87]	0.72 ± 0.02	0.67	0.83	1.0 × 10–4
External validation dataset	0.75 [0.55–0.88]	0.66 ± 0.01	0.68	0.65	0.01
	Subspace Discriminant (A2)
	AUC *	Accuracy	Precision **	Recall **	p ***
Training dataset	0.79 [0.67–0.87]	0.71 ± 0.01	0.69	0.83	0.02
External validation dataset	0.71 [0.52–0.86]	0.63 ± 0.02	0.68	0.65	0.046
Harmo CT features (B)
	Linear SVM (B1)
	AUC	Accuracy	Precision **	Recall **	p ***
Training dataset	0.77 [0.63–0.85]	0.67 ± 0.02	0.74	0.58	1.0 × 10−4
External validation dataset	0.56 [0.39–0.74]	0.58 ± 0.01	0.67	0.52	0.5
	Subspace Discriminant (B2)
	AUC	Accuracy	Precision **	Recall **	p ***
Training dataset	0.76 [0.66–0.87]	0.71 ± 0.02	0.73	0.6	0.01
External validation dataset	0.57 [0.4–0.75]	0.58 ± 0.01	0.67	0.52	0.50
Original CT features (C)
	Linear SVM (C1)
	AUC	Accuracy	Precision **	Recall **
Training dataset	0.56 [0.42–0.68]	0.52 ± 0.03	0.49	0.45
External validation dataset	0.50 [0.34–0.68]	0.43 ± 0.02	0.54	0.65
	Subspace Discriminant (C2)
	AUC	Accuracy	Precision **	Recall **
Training dataset	0.63 [0.48–0.72]	0.56 ± 0.03	0.58	0.56
External validation dataset	0.51 [0.39–0.74]	0.54 ± 0.01	0.58	0.65
PET features only (D)
	Linear SVM (D1)
	AUC	Accuracy	Precision **	Recall **	p ***
Training dataset	0.68 [0.53-0.78]	0.64 ± 0.03	0.64	0.80	0.09
External validation dataset	0.65 [0.43-0.82]	0.64 ± 0.01	0.67	0.78	0.18
	Subspace Discriminant (D2)
	AUC	Accuracy	Precision **	Recall **	p ***
Training dataset	0.71 [0.59–0.82]	0.69 ± 0.01	0.67	0.8	0.10
External validation dataset	0.68 [0.51–0.84]	0.60 ± 0.01	0.67	0.61	0.08
Harmo CT + Original PET + Clinical features (E)
	Linear SVM (E1)
	AUC *	Accuracy	Precision **	Recall **	p ***
Training dataset	0.79 [0.67–0.87]	0.73 ± 0.02	0.72	0.83	6.0 × 10−5
External validation dataset	0.73 [0.54–0.87]	0.73 ± 0.01	0.77	0.74	0.02
	Subspace Discriminant (E2)
	AUC *	Accuracy	Precision **	Recall **	p ***
Training dataset	0.76 [0.65–0.86]	0.74 ± 0.01	0.72	0.83	0.01
External validation dataset	0.75 [0.54–0.88]	0.68 ± 0.02	0.73	0.70	0.02

* AUCs in square brackets are their bootstrapped 95% CIs. ** Precision and recall are presented for class 1. *** p-values are calculated with respect to the conditions C1 and C2 for linear SVM and ESD models, respectively. Values in bold mean the statistical significance.