Fetal Heart Rate Analysis for Automatic Detection of Perinatal Hypoxia Using Normalized Compression Distance and Machine Learning.

Accurate identification of Perinatal Hypoxia from visual inspection of Fetal Heart Rate (FHR) has been shown to have limitations. An automated signal processing method for this purpose needs to deal with time series of different lengths, recording interruptions, and poor quality signal conditions. We propose a new method, robust to those issues, for automated detection of perinatal hypoxia by analyzing the FHR during labor. Our system consists of several stages: (a) time series segmentation; (b) feature extraction from FHR signals, including raw time series, moments, and usual heart rate variability indices; (c) similarity calculation with Normalized Compression Distance, which is the key element for dealing with FHR time series; and (d) a simple classification algorithm for providing the hypoxia detection. We analyzed the proposed system using a database with 32 fetal records (15 controls). Time and frequency domain and moment features had similar performance identifying fetuses with hypoxia. The final system, using the third central moment of the FHR, yielded 92% sensitivity and 85% specificity at 3 h before delivery. Best predictions were obtained in time intervals more distant from delivery, i.e., 4-3 h and 3-2 h.