Figure 1.

Illustration of the study workflow. The photoplethysmogram (PPG) signals were extracted from clinical polysomnographies (PSG), downsampled, and split into three independent sets: training, validation, and test set. These sets were normalized with z-score normalization using the mean and SD of the training set. The signals were then used to generate sequences of hundred 30-s PPG epochs and an overlap of 75% was used in the training set. The sequences were then used to train, optimize, and test the developed neural network resulting in an automatic sleep staging approach utilizing only PPG signal.