Figure 2

Hierarchical clustering to partition the brain activity pattern during the experimental task. Suppose there are \({\varvec{N}}\) brain regions of interests. We can extract brain activation patterns \({{\varvec{X}}}_{{\varvec{i}}}\) during evaluation process for each brain regions. Because each brain regions \({{\varvec{A}}}_{{\varvec{i}}}\) has its own shape and voxel size, however, the voxel intensity itself cannot be compared between brain regions. Hence, representative values (e.g., mean intensity or eigenimage) \({{\varvec{z}}}_{{\varvec{i}}}\) for each sample and brain regions were extracted. The distance (or dissimilarity) \({\varvec{d}}({{\varvec{z}}}_{{\varvec{i}}},{{\varvec{z}}}_{{\varvec{j}}})\) between each brain regions to create a distance matrix. Hierarchical clustering can then be performed using a distance matrix. Note that Pearson’s correlation distance was used as a distance metric in our analysis.