Assessment of non-linear features for intrapartal fetal heart rate classification

Fetal heart rate (fHR) is used to evaluate the fetal well-being during the delivery. It provides information of fetal status and allows doctors to detect ongoing hypoxia. The routine intrapartal evaluation is based on description of macroscopic morphological features of the fHR baseline. FHR contains more information than is used so far, therefore in this work we have focused on evaluation of nonlinear features for fHR signal description. Our data set consists of 189 recordings. Signals with umbilical artery pH less then 7.15 were considered pathological. From each record, 20-minute segment directly preceding the delivery, was chosen. Artifacts were removed from the data and all segments were resampled to 4 Hz sampling frequency. Fractal dimension of attractor, fractal dimension of waveform, entropy, and complexity were used as features. The particular methods used to compute the features were: correlation method for estimation of attractor dimension; Higuchi´s, variance, and box counting method for estimation of waveform fractal dimension; approximate and sample method for estimation of entropy and also the Lempel Ziv complexity was computed. All features were evaluated using Mann-Whitney U test and those fulfilling the statistical significance with p<0.01 were used for further computations. Ten fold cross-validation classification using decision tree and SVM approach was carried out. Overall sensitivity and specificity of 70%, comparable to inter-observer variability, was acquired.