Clustering fetal heart rate tracings by compression

Abstract

Fetal heart rate (FHR) monitoring, before and during labor, is a very important medical practice in the detection of fetuses in danger. We clustered FHR tracings by compression in order to identify abnormal ones. We use a recently introduced approach based on algorithmic information theory, a theoretical, rigorous and well-studied notion of information content in individual objects. The new method can mine patterns in completely different areas, there are no domain-specific parameters to set, and it does not require specific background knowledge. At the highest level the FHR tracings were clustered according to an unanticipated feature, namely the technology used in signal acquisition. At the lower levels all tracings with abnormal or suspicious patterns were clustered together, independent of the technology used. Moreover, FHR tracings with future poor neonatal outcomes were included in the cluster with other suspicious patterns.