Table 2.

Model evaluation and construction. The median and interquartile range is shown over cross-validation splits for SVM models constructed with different imaging modalities. The number of principle components and explained variance thereof refer to automatically tuned hyper-parameters. Permutation test significant *p < .003.

Imaging Modality	Model Evaluation			Model Construction
	Accuracy (%)	Sensitivity (%)	Specificity (%)	Number of Principle Components	Variance of Data Explained by PCs (%)
FA	76.7 [72.4–80.0]*	66.7 [57.1–80.0]	80.0 [76.2–83.3]	17 [9–24]	53.2 [38.2–64.1]
MD	80.0 [75.9–86.2]*	77.8 [66.7–88.9]	82.6 [79.2–86.4]	17 [9–25]	82.3 [70.3–87.8]
AD	82.8 [79.3–86.2]*	80.0 [66.7–88.9]	84.0 [80.0–88.2]	15 [8–24]	77.5 [66.0–85.0]
RD	79.3 [72.4–82.8]*	71.4 [60.0–83.3]	81.0 [77.3–85.7]	18 [9–26]	89.0 [81.2–93.4]
Lesion Segmentation	76.7 [73.3–82.8]*	66.7 [57.1–80.0]	81.8 [77.3–85.0]	15 [7–23]	80.7 [65.5–89.2]