Individual-Level Prediction of Exposure Therapy Outcome Using Structural and Functional MRI Data in Spider Phobia: A Machine-Learning Study
Table 2
Prediction results of the 1st-level classifiers of posttreatment response ( responders vs. nonresponders).
1st-level classifier
Balanced accuracy (SD)
AUROC (SD)
Functional features extracted from a priori selected 30 ROIs
Demographic and questionnaire data
0.60 (0.07)
0.64 (0.08)
Structural MRI
0.51 (0.07)
0.51 (0.08)
Functional activation
0.48 (0.08)
0.48 (0.09)
Phasic fear gPPI
0.52 (0.06)
0.53 (0.08)
Sustained fear gPPI
0.51 (0.06)
0.55 (0.08)
Phasic fear graph measures
0.52 (0.07)
0.54 (0.08)
Sustained fear graph measures
0.54 (0.07)
0.56 (0.09)
BOLD variance
0.55 (0.07)
0.59 (0.08)
Functional features extracted from ROIs across the whole brain (exploratory)
Demographic and questionnaire data
0.60 (0.07)
0.64 (0.08)
Structural MRI
0.51 (0.07)
0.51 (0.08)
Functional activation
0.51 (0.06)
0.52 (0.08)
Phasic fear gPPI
0.49 (0.05)
0.49 (0.10)
Sustained fear gPPI
0.49 (0.05)
0.49 (0.09)
Phasic fear graph measures
0.48 (0.06)
0.50 (0.07)
Sustained fear graph measures
0.47 (0.06)
0.49 (0.08)
BOLD variance
0.63 (0.07)
0.67 (0.08)
ROI: region of interest; AUROC: area under the receiving operator curve; gPPI: generalized psychophysiological interaction. , corrected resampled -test against the accuracy of a dummy classifier, two-tailed.
