Figure 3.

The pipeline of converting functional near-infrared spectroscopy (fNIRS) signals to brain functional networks (BFNs) (only take the first subject in the dataset as an example). The size of the original Oxy-Hb data is 30003*20, where the 30003 is the number of time points, and the 20 is the number of channels. By segmenting the original data according to the phase of task and resting state for each trial, we obtain 150 data matrices, half of which have a size of 133*20 under task state, and the remaining have a size of 200*20 under resting state. Note that, when conducting the segmentation operation, we truncate the time points that lie at task phase and the intermediate of the rest phase, respectively, for ensuring signal purity and eliminating the signal noise caused by task-resting switching. The data matrix is then used to estimate BFN fingerprint by PC. As a result, the subject obtains 150 BFNs, each of which has a size of 20*20. All BFNs are classified into four classes according to four tasks (i.e., RHT, LHT, FT, and REST). Each task class contains 25 BFNs and the resting class contains 75 BFNs. Since the size of each BFN is same, the subject would obtain a new BFN by averaging the element of the corresponding position of all BFNs within one class. Eventually each subject obtains 4 BFNs corresponding to 4 tasks.