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

Cardiotocography is a noninvasive method that is used to evaluate fetal health status in antepartum and intrapartum tracking. A typical cardiotocogram recording consists of two distinct signals, namely, fetal heart rate (FHR) and uterine activity. However, FHR signals suffer from invalid or missing samples due to artifacts or sensor malfunction. These missing samples disrupt signal continuity and result in inadequate characteristics for the FHR signal. This deficiency affects the precision of follow-up analysis. Various techniques have been proposed to improve the evaluation of short dropouts, but they tend to be problematic when used for long dropouts. Furthermore, traditional pointwise metrics have a number of limitations when applied to FHR analysis. The over-reliance on traditional metrics to recover full sample points presents certain risks. Hence, we prefer to restore the missing heart rate patterns in FHR signals. This study proposes an imputation technique based on optimal transport theory and a new metric to evaluate the technique’s prediction performance. Experiments on a batch of public data sets show that the imputation accuracy and robustness of the proposed technique outperforms those of existing standard methods.