50. Comparison of canonical correlation analysis and pearson correlation in resting state FMRI in patients with parkinson’s disease

Abstract

Introduction In many fMRI studies of functional connectivity, researchers need to select representative signal of region of interest (ROI). Usual approach is to compute mean or first eigenvector of ROI voxels and use it to calculate Pearson correlation (PC). In this abstract we compare this usual practice with extended technique using more eigenvectors and canonical correlation analysis (CCA). Methods We used resting-state data from 18 PD patients, two sessions for each subject, first in off state, second in on state (after l -dopa medication). Dataset was acquired using 1.5T MR scanner Siemens Symphony, 150 scans, TR = 3.0 s. Data were preprocessed in SPM8 – unwarped, slice timing corrected, spatially normalized and smoothed using 5 mm kernel. Masks of individually segmented gray matter were applied. Then we used AAL parcelation and performed PCA on each region. First eigenvectors were used for PC analysis and every eigenvector above 10% explained variability entered CCA. To find AAL regions influenced by medication, t-tests were performed on differences between first and second session both using PC and CCA coefficients. Results CCA identified 107 and PC 206 significant changes in correlation between AAL regions (p Conclusion PC and CCA are statistical methods quantifying information about similarity of signals. Although CCA is not commonly used in fMRI, we performed it as extension of PC, because we want to exploit more variability from data than PC can. Mean CCA correlation on group level is significantly higher than PC correlation. CCA evaluates greatest correlation obtained by linear combination of eigenvectors between every two regions. Therefore CCA contains stronger correlations than PC. Statistically significant results are evidence that CCA is able to find differences between datasets albeit on uncorrected level. These differences can represent effect of l -dopa medication on functional connectivity network. CCA could reveal information, that remains hidden for PC analysis and for assessment of functional connectivity it could be useful to combine information from both methods – CCA and PC. Thanks to Grant GA14-33143S from GACR for funding.