Analyzing Fetal Heart Rate Patterns via Latent Representations with Sequential Variational AutoEncoder (SeqVAE)

Fetal heart rate (FHR) monitoring is a non-invasive method for assessing fetal well-being during labor, providing vital insights into potential fetal distress and the risk of hypoxic-ischemic encephalopathy (HIE).However, traditional clinical methods for interpreting FHR signals are often subjective.In this paper, we propose a novel Sequential Variational Autoencoder (SeqVAE) model, inspired by the Variational Recurrent Neural Network (VRNN), to learn latent representations of FHR signals.The SeqVAE model is designed to capture the temporal dependencies within the FHR signals while maintaining a probabilistic framework, resulting in smoother and more informative latent representations compared to the VRNN.These latent representations are then utilized for the classification of FHR signals into healthy and HIE cases.Our experiments demonstrate that the SeqVAE model outperforms the VRNN in generating latent representations that align closely with significant physiological events in FHR signals.