Table 2.

Performance results of the classifiers

Model category	Model name	FPR	TPR	F0.5 score†	ROC/ AUC a	Recallb	Precisionb	Kappa  (-1, 1)
Approach #1: Clinical data only (on the set of 30 selected clinical labels from ExtraTree classifier)	Gaussian NB	0.49	0.70	0.57	0.60	0.61	0.59	0.19
	Random Forest	0.07	0.56	0.60	0.74	0.74	0.64	0.27
	Gradient Boosting	0.14	0.70	0.66	0.78	0.78	0.68	0.37
	XGBRF	0.16	0.65	0.62	0.72	0.73	0.64	0.28
	k-nearest neighbors	0.47	0.58	0.50	0.55	0.56	0.52	0.05
	SVM	0.18	0.18	0.32	0.50	0.50	0.42	0
	MLP	0.40	0.40	0.22	0.50	0.50	0.20	0
Approach #2: CTs only	3D-CNN	0.83	0.84	0.63	0.57	0.57	0.78	0.22
	3D Swin Transformer	0.38	0.89	0.75	0.75	0.75	0.75	0.49
Approach #3: Data fusion	Terminal 3D-CNN on CTs + 30 labels	0.36	0.90	0.75	0.76	0.76	0.76	0.51
	Medial 3D-CNN on CTs + 67 labels	0.65	0.98	0.70	0.66	0.66	0.67	0.37
	3D Swin Transformer on CTs + 30 labels	0.35	0.91	0.78	0.78	0.78	0.80	0.55
	3D Swin Transformer on CTs + 67 labels	0.40	0.95	0.82	0.77	0.77	0.83	0.60

^aROC Receiver Operating Characteristics Curve, AUC Area under the ROC Curve

^bDenotes the macro-averaging evaluation method