Evaluation of fetal heart rate recordings based on clustering

Fetal heart rate (FHR) recorded within the cardiotocography (CTG) measurement is currently the main method to evaluate fetal health state during delivery. The CTG provides valuable information about fetal behavior as a reaction to stressful events (hypoxic episodes). The presented paper proposes to use data driven analysis of FHR - the clustering analysis of features derived automatically from the signal using novel method of signal approximation called SAX Even though the clustering is well grounded in signal processing tasks in the field of FHR research it is used sparingly due to high inter-individual variability of fetuses and dificulties to link diferent temporal events. Data from the open access CTU-UHB database (552 CTG records) available at the Physionet are used and the 30 minutes segments at the end of the first stage of labor are analyzed. The classification based on clustering achieved sensitivity of 58. 7% and specificity of 69.4% for the two class classification - well comparable to other pH based studies. Sensitivity was improved to 71.4% for six cluster settings - thus suggesting different classes of FHR. This is in contrast with objective evaluation (two classes usually determined with pH threshold) and three classes (used by clinicians for evaluation). Nevertheless to properly describe the link between these clusters and clinical evaluation robust interpretation is still necessary.