Clinicians often rely upon the cardiotocogram, a display of the fetal heart rate and maternal uterine activity (UA) over time, as a means of monitoring fetal health during labour. Fetal health can be monitored adequately only when the signal quality of the cardiotocogram is good. We propose an automated assessment of UA signal quality in order to create a confidence index for subsequent analysis of the intrapartum cardiotocogram. We use an autoregressive (AR) model of the UA to estimate the power at the contraction frequency, with high power indicative of "good" UA signal quality. 5th, 10th, and 15th-order AR models are used to assess the signal quality of 12 intrapartum UA traces as "good/medium" or "poor". We compare our results to two experts' visual assessments of signal quality. The 10th-order model exhibits the highest percent agreement rate of 62%. It also exhibits the most balanced false positive and false negative rates, where "good" or "medium" signal quality is considered a positive and "poor" signal quality a negative. The 10th-order model can therefore be used as a confidence index to reduce the errors made in the identification of uterine contractions in the UA trace and in the subsequent analysis of the cardiotocogram as a whole.

Original publication

DOI

10.1016/s1350-4533(01)00092-3

Type

Journal article

Journal

Medical engineering & physics

Publication Date

11/2001

Volume

23

Pages

603 - 614

Addresses

Signal Processing and Neural Networks, Department of Engineering Science, University of Oxford, Parks Road, Oxford OX1 3PJ, UK. shelley@robots.ox.ac.uk

Keywords

Humans, Fetal Distress, Cardiotocography, Models, Statistical, Pregnancy, Uterine Contraction, Heart Rate, Fetal, Signal Processing, Computer-Assisted, Female