A non-parametric model for Ballistocardiography

In this paper we propose a probabilistic generative model for the Ballistocardiogram (BCG), a physiological signal derived from the recoil of the body caused by the beating heart. The model uses a Gaussian process for the continuous BCG signal and an inverse Gaussian point process to model the latent discrete heartbeat sequence. Using this model artificial BCGs can be generated for the purpose of validating BCG analysis methods or to estimate missing data. We also demonstrate how accurate heartbeat estimates can be inferred from real BCGs by employing Markov chain Monte Carlo.