Figure 2.

Schematic pipeline of the proposed EEG pre-processing strategy for seizure prediction: (a) EEG-to-scalogram conversion procedure: continuous wavelet transform (CWT) is adopted to generate the EEG power spectrum from the time-series EEG data; and 3D-to-2D projection (Proj) is used to produce the 2D time-frequency representations of EEG named “scalogram”. (b) EEG pre-processing approach: $S_1$, $S_2$, ⋯, $S_{60}$ correspond to the 1st, 2nd, and 60th 10-s segments of each 10-min EEG clip ($f_S$ = 400 Hz); N is the total number of EEG channels (N = 23 for scalp EEG; N = 16 for invasive EEG); d is the number of data-points in each EEG segment (d = 10-s ×$f_S$ = 4000); and h and w are the height and width of the EEG scalogram images ($h\times w$ = 100 × 4000).