Table 1.
Measurements of the modulation profiles in sPCA+CCA, PCA+CCA, parallel ICA and sCCA.
| Methods | ECM | VBM | Correlation ρ | Classification accuracy | ||
|---|---|---|---|---|---|---|
| pcorr | AUC | pcorr | AUC | |||
| sPCA+CCA | 3.4 × 10−4 | 0.78 | 2.6 × 10−4 | 0.81 | 0.78 | 0.68 |
| PCA+CCA | 2.9 × 10−2 | 0.71 | 2.0 × 10−3 | 0.73 | 0.48 | 0.61 |
| Parallel ICA | 0.10 | 0.68 | 1.8 × 10−3 | 0.75 | 0.27 | 0.58 |
| sCCA | 1.3 × 10−2 | 0.70 | 1.6 × 10−2 | 0.70 | 0.80 | 0.60 |
sCCA obtains the highest correlation due to overfitting, but has very low values in AUC. sPCA+CCA has largest AUC and a large correlation indicating that this method is superior to the other 3 methods. pcorr denotes the p value for group discrimination with multiple comparison correction. The group classification accuracy is obtained by running quadratic discriminant analysis on modulation profiles from both modalities. The most significant p-value and the largest value for each measurement are in bold font.