Extending Local Canonical Correlation Analysis to Handle General Linear Contrasts for fMRI Data

Figure 2

Determination of the proper value for the noise fraction 𝑓 for pseudoreal data. The solid horizontal lines represent the medians of 𝑡-statistics at the significance level 10⁻⁸<𝑃<10⁻³ (uncorrected) for the contrasts V-F (a) and E-C (b) by applying GLM-NS on real fMRI activation data. The dashed curves are the medians of the 𝑡-statistics of activation-defined voxels for the contrasts V-F (a) and E-C (b) by applying GLM-NS on pseudoreal data where the true activations were defined by thresholding real activation data using a very high significance level (𝑃<10⁻⁸ uncorrected) and adding resampled noise according to (16) for different noise fractions𝑓∈[0,1]. The medians of the 𝑡-value matched at around 𝑓=0.6. Therefore, we picked two values for 𝑓∶𝑓=0.55 representing the low noise case and 𝑓=0.65 representing the high noise case.

(a) Visual

(b) Memory