Item-Wise Interindividual Brain-Behavior Correlation in Task Neuroimaging Analysis

Brain-behavior correlations are commonly used to explore the associations between the brain and human behavior in cognitive neuroscience studies. There are many critics of the correlation approach, however. Most problems associated with correlation approaches originate in the weak statistical power of traditional correlation procedures (i.e., the mean-wise interindividual brain-behavior correlation). This paper proposes a new correlation procedure, the item-wise interindividual brain-behavior correlation, to greatly enhance statistical power by testing the significance of small correlation coefficients from trials against zero rather than simply pursuing the highest possible correlation coefficient. The item-wise and mean-wise correlation were compared in simulations and the data analyses for a task fMRI experiment on mathematical problem-solving. Stimulations show that item-wise correlation relative to mean-wise correlation results in smaller r value when data are random and higher t value when signal-to-noise ratio is equal to or larger than 6%. Item-wise correlation displayed more voxels with significant brain-behavior correlation than did mean-wise correlation. The analysis with item-wise correlation other than with mean-wise correlation shows significant brain-behavior correlation at the threshold of p<.05 corrected. Cross validation showed that odd- and even-ordered trials had greater stability in correlation for item-wise correlation (r=.918) than for mean-wise correlation (r=.686). All stimulations and example analyses demonstrate the effectiveness of the proposed correlation procedure for task neuroimaging studies.