Stability of variability features computed from fetal heart rate with artifcially infused missing data

Fetal heart rate (FHR) is usually the only measurement acquired directly from the fetus during delivery; therefore, any distortion leading to a corrupted signal means difficulties for fetal surveillance. Obstetricians use morphological features for evaluation of FHR that are usually estimated visually. Other type of features are those that are either difficult to estimate visually or can not be estimated by naked eye at all. In this paper we examined several STV (short), LTV (long term variability) features and one nonlinear feature. We compared features for an undistorted and distorted signal with missing samples. We analyzed feature's stability with respect to acquisition technique (ultrasound measurement, direct ECG measurement). The stability of STV indicies were almost identical except standard STV, De Haan index, and Yeh index. These indices changed markedly with increasing percentage of missing samples; Yeh index decreased up to 70% of its nominal value and De Haan index increased up to 350%; this behavior held for FHR recorded using ultrasound measurement and even for FHR directly computed from fetal ECG. This relationship was present for beat-to-beat variability and disappeared for epoch-to-epoch variability.