Table 2.

Performance comparison of the ML techniques using either features from the delineated tumor (D) or from the rough VOI (V), in addition to the available clinical factors.

ML	Task	VOIa	Training set			No. of features	Test set
			Se	Sp	BAcc		Se	Sp	BAcc
LR	Median OS	D	0.67	0.77	0.72	37	0.54	0.75	0.63
		V	0.58	0.68	0.63	24	0.59	0.57	0.58
	6-month OS	D	0.81	0.87	0.84	45	0.8	0.76	0.78
		V	0.74	0.78	0.76	32	0.61	0.65	0.63
RF	Median OS	D	0.87	0.91	0.89	25	0.60	0.75	0.67
		V	0.75	0.86	0.87	23	0,53	0.59	0.56
	6-month OS	D	1	1	1	47	0.74	0.86	0.80
		V	0.83	0.89	0.86	58	0.73	0.75	0.74
SVM	Median OS	D	1	1	1	27	0.53	0.73	0.64
		V	0.82	0.82	0.82	20	0.56	0.60	0.58
	6-month OS	D	0.88	0.96	0.92	38	0.76	0.74	0.75
		V	0.84	0.90	0.87	43	0.75	0.77	0.76
Fusion (average of output probabilities)	Median OS	D	1	1	1	-	0.76	0.80	0.78
		V	0.93	0.89	0.90	-	0.76	0.78	0.77
	6-month OS	D	1	1	1	-	0.91	0.87	0.89
		V	0.88	0.94	0.91	-	0.98	0.78	0.88

ML, machine learning; VOI, volume of interest; Se, sensitivity; Sp, specificity; BAcc, balanced accuracy; LR, logistic regression; RF, random forest; SVM, support vector machine.

^aD stands for the accurately delineated tumor and V for the “rough” VOI.