FIGURE 1.

Architecture overview of the autonomic arousal detection model. Panel (A) provides an overview over the architecture of the autonomic arousal detection model developed in the present study. There are two feature extraction modules that extract cardiac features from the photoplethysmography signal and respiratory features from the respiratory flow signal. The cardiac and respiratory features are combined and used as input for the final arousal detection module which generates an arousal probability sampled at 2 Hz. All parts of the model use residual convolutional network blocks as illustrated in panel (B) to successively increase the feature complexity while reducing temporal resolution.