Recovery of missing samples in fetal heart rate recordings with Gaussian processes

Abstract

Abstract-Missing samples are very common in fetal heart rate (FHR) recordings due to various reasons including fetal or maternal movements and misplaced electrodes. They introduce distortions and cause difficulties in their analysis. In this paper, we propose a Gaussian process-based method that can utilize other intrapartum signals (e.g., uterine activity and maternal heart rate) to recover the missing samples in FHR recordings. The proposed approach was tested on a short real FHR recording segment and its performance was compared with that of cubic spline interpolation which is widely used in pre-processing of FHR recordings. Our results show that the proposed approach, with utilization of UA signals, achieves 2.35 dB to 14.85 dB better recovery performance. Furthermore, even when the percentage of missing samples is more than 50%, the mean square error of this approach is still below one beat per minute.

Inference about Causality from Cardiotocography Signals Using Gaussian Processes

“…With this information, the predictive performance of the method should be better than that of using information from past values of FHR alone. This is consistent with our observation in [19]. There we showed that adopting information from UA signals helps in recovering missing samples of FHR tracings.…”

“…GPs have been successfully applied in both supervised and unsupervised learning tasks [18]. For example, in our previous work [19], we proposed a GP-based method that employs UA signals to recover missing samples of FHR recordings and had excellent results. This work also provided evidence that the UA signals contain information about fetal well-being.…”

Imputation Method for Fetal Heart Rate Signal Evaluation Based on Optimal Transport Theory

“…MAE, RMSE, and W 2 imputation metrics are defined as Different methods of imputation performance were compared in four standard FHR features: the Approximate Entropy (ApEn) [29], Short-Term Variability (STV) [7], Long-Term Variability (LTV), and Δ (Delta) [8]. The formula is defined as follows:…”

“…The preprocessing step is vital in FHR evaluation, because its quality influences the values of features and, consequently, the evaluation's performance [5][6][7]. For instance, [8] explored the quality of several STV and LTV functions when 0-50% of samples are missing; the functions were randomly picked in a 5 min FHR segment within the initial stage of labor. These missing samples were then linearly interpolated before analysis.…”

Signal Processing for Newborn Survival from labour to resuscitation

“…As measurements of FHR using Doppler ultrasound is susceptible to signal dropouts, described in section 2.1.2, any missing data should be estimated prior to any automated analyses. Simple methods such as linear interpolation [61] and cubic Hermite spline interpolation [62], and more complex methods such as Gaussian processes [63] and K-SVD [29] have previously been used to estimate the missing samples on FHR recorded through CTG. However, depending on the length of the gaps, these methods can affect computation of the traditional heart rate features such as variability.…”