FIGURE 2.

Structure of the CNN (A) and the CNN-LSTM (B) networks. Input is on top, output at the bottom. Since we applied several configurations of the CNN networks the repetitions of the last convolutional and pooling blocks were different. The number of channels in the input may differ for the networks using either an EEG and two EOG channels or a single EEG channel only. See sections “Architecture of the Networks” and “Materials and Methods” and Table 2 for the description of the different layers.