Table 10.
Meta-dataset and feature type | Harmonization efficacy (age-group permutation test p-value, One-sided Wilcoxon signed-rank test p-value) | |
---|---|---|
Harmonization with neuroHarmonize | Harmonization with harmonizer transformer | |
CHILDHOOD | ||
CT | (0.5363, 10−18) | (0.0036, 10−18) |
FD | (0.5853, 10−18) | (0.0268, 10−18) |
ADOLESCENCE | ADOLESCENCE | |
CT | (0.3559, 10−18) | (0.0484, 10−18) |
FD | (0.0090, 10−18) | (0.0002, 10−18) |
ADULTHOOD | ADULTHOOD | |
CT | (0.4545, 10−18) | (0.4727, 10−18) |
FD | (0.0042, 10−18) | (0.0006, 10−18) |
LIFESPAN | ||
CT | (0.1128, 10−18) | (0.0006, 10−18) |
FD | (0.0002, 10−18) | (0.0002, 10−18) |
The age-group permutation test p-value and one-sided Wilcoxon signed-rank test p-value have been reported. The permutation test p-value indicates whether the site effect has been removed (i.e., p-value ≥ 0.05 means that the imaging site prediction is not different from a random prediction). One-sided Wilcoxon signed-rank test p-value indicates whether the site effect has been reduced (i.e., p-value less than 0.05 means that the prediction of imaging site using the harmonized features obtains a balanced accuracy significantly less than that estimated using raw data).
CT: cortical thickness; FD: fractal dimension.