A respiratory latent variable model for mechanically measured heartbeats.

The effect of respiration on heartbeats in a mechanically measured cardiac signal was noticed already in the first respiratory ballistocardiography studies in the 1920s. The effect was described as a modulation of the heartbeat amplitude, and that view has persisted. Although a reasonable approximation, amplitude modulation is not an accurate description of the respiratory effect. In this paper, the respiratory variation is described with a linear latent variable model, where the direction of variation can be distinct from the direction of amplitude. The model was evaluated with seven ballistocardiograms from three healthy test subjects. Based on a Bayesian information criterion analysis, it was found to describe the variation accurately in all the cases. As the modelling of the heartbeat shape is improved, existing heartbeat detection methods can be made more accurate by applying the proposed model.