Table 1:

Stability of machine learning results calculated independently in test (Scan1) and retest (Scan2) scans. (A) Feature pruning method. (B) Selection of 10 best ranking features according to pruning method. (C) Median and range of repeatability of features used in classifier training. (D) Agreement of binary classifications from classifiers trained with Scan1 and Scan2 data. (E) Area Under Receiver operator characteristic curve (AUC) when classifications are evaluated with Scan1 of test set. (F) AUC with Scan2. (G) Pooled estimate over Scan1 and Scan2 for AUC.

Pruning method (A)	Pruned features (B)	ICC median (range) (C)	% of same labels (D)	AUC (95% CI) Scan1 (E)	AUC (95% CI) Scan 2 (F)	AUC (95% CI) Scan 1 and 2 (G)
Wilcoxon rank-sum test	ADCk Top 10	0.438(0.000..0.953) 0.280(0.000..0.693)	80.4	0.682 (0.490..0.873)	0.495 (0.261..0.730)	0.586 (0.436..0.737)
MRMR	ADCk Top 10	0.296(0.000..0.813) 0.233(0.000..0.621)	70.6	0.673 (0.469..0.876)	0.630 (0.438..0.821)	0.649 (0.514..0.785)
Spearman ρ	ADCk Top 10	0.471(0.445..0.754) 0.476(0.414..0.654)	70.6	0.743 (0.558..0.928)	0.645 (0.465..0.826)	0.691 (0.563..0.819)
AUC	ADCk Top 10	0.471(0.445..0.754) 0.476(0.414..0.654)	70.6	0.743 (0.558..0.928)	0.645 (0.465..0.826)	0.691 (0.563..0.819)
Wilcoxon rank-sum test	ADCk ICC>0.8 & Top 10	0.823(0.804..0.953) 0.823(0.804..0.953)	84.3	0.783 (0.637..0.929)	0.750 (0.592..0.908)	0.770 (0.664..0.875)
MRMR	ADCk ICC>0.8 & Top 10	0.888(0.808..0.962) 0.836(0.800..0.962)	74.5	0.727 (0.538..0.916)	0.732 (0.566..0.897)	0.730 (0.607..0.852)
Spearman ρ	ADCk ICC>0.8 & Top 10	0.904(0.800..0.951) 0.927(0.800..0.942)	72.5	0.686 (0.503..0.870)	0.723 (0.566..0.880)	0.706 (0.589..0.822)
AUC	ADCk ICC>0.8 & Top 10	0.904(0.800..0.951) 0.927(0.800..0.942)	72.5	0.686 (0.503..0.870)	0.723 (0.566..0.880)	0.706 (0.589..0.822)