KCCA for fMRI analysis

2 Results As it is impossible to tell which method is better when real data is used, we experiment with controlled simulated data. We embed square-wave “activity” in a nulldata set (no brain activity). The paradigm of the applied activity is 10 images rest, 10 images activity and so forth. Resulting with 200 time points. As we know the activation period we use for our time sequence a square-wave representation of activity 1 and rest− 1 over the 200 time-course. We find that when plotting the precision vs. recall values for the pixels identified by KCCA and CCA. KCCA is consistently better then the baseline method. Splitting the data into training and testing we are able to reconstruct the signal of the fMRI test-set with an average of 97.4% accuracy over 100 random repeats.