. 2016 Oct 24;3(4):047502. doi: 10.1117/1.JMI.3.4.047502

Table 5.

Mean and standard deviation of AUC and accuracy of QH features in distinguishing tumor from benign regions over 100 iterations of a random forest classifier using Wilcoxon rank-sum test for feature selection at the $p < 0.05$ confidence level. Intrasite classification performed using threefold cross validation, intersite classification performed using independent validation. Reported as average of $D_{1}$ and $D_{2}$ results. No texture features were predictive, and the family is therefore not represented here.

	Graph	Shape	CGT	Subgraph	All families
Intrasite AUC	$0.89 \pm 0.12$	$0.96 \pm 0.03$	$0.91 \pm 0.07$	$0.90 \pm 0.10$	$0.93 \pm 0.05$
Intersite AUC	$0.83 \pm 0.09$	$0.97 \pm 0.02$	$0.80 \pm 0.02$	$0.84 \pm 0.04$	$0.95 \pm 0.03$
$p$ -Value	$5.7 \times 10^{- 12}$	0.88	$1.3 \times 10^{- 52}$	$2.0 \times 10^{- 8}$	0.29
Intrasite accuracy	$0.87 \pm 0.15$	$0.89 \pm 0.06$	$0.88 \pm 0.06$	$0.88 \pm 0.12$	$0.86 \pm 0.09$
Intersite accuracy	$0.75 \pm 0.05$	$0.90 \pm 0.04$	$0.73 \pm 0.03$	$0.75 \pm 0.05$	$0.84 \pm 0.09$
$p$ -Value	$1.6 \times 10^{- 19}$	$3.6 \times 10^{- 4}$	$6.9 \times 10^{- 62}$	$5.5 \times 10^{- 41}$	0.01