Figure 2. .

Performance on individual identification regarding the number of neighbors and the number of input frames. The highest mean classification accuracy was achieved with k = 5. With 20 frames or fewer as input data, cGCN achieved significantly higher classification accuracy compared with random GCNs and the convolutional RNN (ConvRNN) model in our prior study (L. Wang et al., 2019). Particularly with 5 frames, cGCNs obtained the identifying accuracy of over 49% with k = 5, which was much higher than random GCNs and ConvRNN. It demonstrated that spatiotemporal features from each frame of fMRI data was successfully extracted by cGCNs for the individual identification task.