Skip to main content
Biophysics

Biophysics

Improving the detection of foetal distress

21 Mar 2018 Tami Freeman
HRV features measured during rest periods and contractions
HRV features measured during rest periods and contractions

The health of a foetus is monitored during labour to check for signs of foetal distress. Currently, this is performed using cardiotocography (CTG), a technique that monitors the foetal heart rate (FHR) and uterine activity. CTG is interpreted visually, however, and suffers from high inter- and intra-observer variability and low specificity.

Foetal heart rate variability (HRV) can also provide information on foetal distress, but it is strongly influenced by uterine contractions, particularly during the second stage of labour. A Dutch research team is now investigating whether measuring HRV separately during contractions and rest periods can improve detection of foetal distress (Physiol. Meas. 39 025008).

“In our study, we focused on foetal distress related to oxygen deficiency,” explained Guy Warmerdam from Eindhoven University of Technology. “During labour, uterine contractions can temporarily block of oxygen supply to the foetus. Normally, the foetus can handle such stress well. However, if oxygen deficiency is severe and prolonged this can lead to oxygen deficiency in the central organs of the foetus, potentially damaging them.”

Case comparisons
The researchers studied FHR signals from 20 cases with adverse foetal outcome and 80 healthy cases. Foetal outcome was based on the acid-base balance in the foetal blood after birth (which is related to the oxygen concentration in the foetal blood), with an adverse outcome defined as a pH below 7.05 and healthy as above 7.20. As the effects of oxygen deficiency increase as labour progresses, they only considered FHR segments recorded up to a maximum of 45 minutes before birth.

For each 10 minute FHR segment, they calculated a series of HRV features. These included standard deviation (SD) and root mean square of successive differences (RMSSD), sample entropy (SampEn), a scaling exponent (α) and deceleration capacity (DC) – the average response to a deceleration in heart rate. Using spectral analysis of the FHR, they also calculated power in the low-frequency (LF) and high-frequency (HF) bands, total power (TP) and normalized frequency powers LFn and HFn.

In addition to examining the entire FHR segment, the researchers also calculated HRV features separately during contractions and rest periods and determined the ratios between these. As the length of a contraction or rest period was often less than one minute, this analysis was limited to four features: SD, RMSSD, HF and SampEn. They observed that these features were all higher during contractions than during rest periods.

To select the best combination of HRV features to detect foetal distress, the researchers employed a genetic algorithm (GA). For this, they used the 10 minute FHR segment closest to birth for each foetus, and examined three data sets: S1, HRV features calculated over the entire FHR; S2, only contraction-dependent features; and S3, the combination of features from S1 and S2. Due to the relatively small dataset, they repeated the GA process 50 times using different data splits, generating 50 subsets of HRV features for S1, S2 and S3.

The HRV features most frequently selected by the GA were TP, LFn, HFn and DC for S1; SDratio and RMSSDuc for S2; and SDratio and HF for S3. The top most selected feature was TP for S1, and SDratio for S2 and S3. Both TP and SDratio are related to the presence of heart rate decelerations. SDratio also contains information about how the foetus recovers from contractions: a high value indicates that the foetus recovers quickly and stabilizes its cardiovascular system during rest periods, while a low ratio indicates that the foetus is unable to recover.

The researchers used support vector machines (SVMs) to determine the classification performance of these HRV feature sets. They trained the classifier using the geometric mean (g), which represents a balance between classification accuracy of the healthy majority class (specificity) and the minority class (sensitivity). The average cross-validation performance, g, for classification of FHR segments closest to birth improved from 70% for S1 (which did not include contraction-dependent data), to 76% for S2 and 79% for S3.

Early intervention
The earlier that foetal distress can be predicted, the more useful the information for clinical intervention. Thus, the researchers examined classification performance over time. They trained a classifier for each feature subset using the FHR segments closest to birth, then used the trained classifiers to determine foetal distress for all segments from 45 minutes before until the time of birth.

For all three data sets, the classification performance, g, increased towards the time of birth. The sensitivity also increased nearer to the birth time, while the specificity decreased. At 15 minutes before birth, g was 60% for S1, 69% for S2 and 72% for S3.

“This work showed that separating contractions from rest periods improves HRV analysis for the detection of foetal distress during labour,” said Warmerdam. “The dataset we used contained a relative small number of foetuses with poor outcome; a larger dataset is required to gain more insight into which combination of features works best.”

“In this study, we focused on the second stage of labour, the stage of active pushing. It would be interesting to examine the performance during the first stage of labour,” he told medicalphysicsweb. “Finally, our study was limited to binary classification (good versus bad outcome). Before a classifier can be used in clinical practice, a future study should use a dataset containing all foetal outcomes.”

Copyright © 2024 by IOP Publishing Ltd and individual contributors