Figure 3. The latent dimensions can be organized into 6 groups (shown in rows) based on their spatial similarities. Panel A shows how the spatial profile at the max-variance time (in figure 2) changes when sliding a single latent variable from -3 to +3. Panel B shows the spatial similarities among latent variables, which were clustered using K-means clustering (K = 6). The cluster number and the variance explained were also shown. Panel C shows the weighted mean functional connectivity of each cluster of latent variables.

Figure 2. Spatial temporal patterns extracted by latent dimensions. Each subplot is obtained by making one latent variable equal to +3 (+3σ for Gaussian distribution) while fixing the rest of the latent variables at zero. The x-axis is time in seconds. The y-axis is the 246 parcels. The patterns have arbitrary units, but all subplots share the same display scale so that higher variance results in higher contrast. The 32 latent variables are already organized in 6 clusters (see figure 3). The black cursor indicates the maximum variance time.