Table 3:

Model accuracy—accuracy for random forest models trained on conventional (anatomic imaging) and conventional-plus-advanced (diffusion, perfusion, permeability) imaging^{^a}

Metric	Variables	Conventional	Conventional-Plus-Advanced
Accuracy	All Variables	0.904	0.952
	RF Rank	0.875	0.933
	RF (fixed variables)	0.885 (T2, T1C, FLAIR, T1)	0.962 (T2, ADC, CBV, K^trans)
κ	All Variables	0.822	0.912
	RF Rank	0.768	0.877
	RF (fixed variables)	0.788	0.930

Note:—RF indicates random forest.
↵a Performance using all 23 variables is compared with the performance averaged over 5 folds of cross-validation as well as the results using a fixed-variable subset. The Cohen κ is larger for more accurate models but is scaled so that a no-information classifier would have κ equal to zero.