Optimization of Imminent Labor Prediction Systems in Women with Threatened Preterm Labor Based on Electrohysterography

Abstract

Preterm birth is the leading cause of death in newborns and the survivors are prone to health complications. Threatened preterm labor (TPL) is the most common cause of hospitalization in the second half of pregnancy. The current methods used in clinical practice to diagnose preterm labor, the Bishop score or cervical length, have high negative predictive values but not positive ones. In this work we analyzed the performance of computationally efficient classification algorithms, based on electrohysterographic recordings (EHG), such as random forest (RF), extreme learning machine (ELM) and K-nearest neighbors (KNN) for imminent labor (<7 days) prediction in women with TPL, using the 50th or 10th-90th percentiles of temporal, spectral and nonlinear EHG parameters with and without obstetric data inputs. Two criteria were assessed for the classifier design: F1-score and sensitivity. RF_{F1_2} and ELM_{F1_2} provided the highest F1-score values in the validation dataset, (88.17 ± 8.34% and 90.2 ± 4.43%) with the 50th percentile of EHG and obstetric inputs. ELM_{F1_2} outperformed RF_{F1_2} in sensitivity, being similar to those of ELM_Sens (sensitivity optimization). The 10th-90th percentiles did not provide a significant improvement over the 50th percentile. KNN performance was highly sensitive to the input dataset, with a high generalization capability.

Keywords: K-nearest neighbors; electrohysterogram; extreme learning machine; imminent labor prediction; random forest; tocolytic therapy; uterine myoelectrical activity.

Figure 1

Scheme of the method used to train, validate and test the imminent labor prediction classifiers (time to delivery (TTD ≤ 7) based on EHG in women with threatened preterm labor. This was performed with two optimization criteria in the classifier design: F1-score and sensitivity.

Figure 2

Mean values of different RF classifier metrics for validation datasets in the 30 data partitions optimized by F1-score. The same results were obtained when optimizing by sensitivity. For each metric the significant differences (p < 0.05) for each input dataset are marked with: 10th–90th percentiles of EHG parameters + obstetric input data; 50th percentile of EHG + obstetric input data; 10th–90th percentiles of EHG parameters; 50th percentile of EHG parameters.

Figure 3

Mean values of different ELM classifier metrics for validation datasets in the 30 data partitions (a) optimizing F1-score (b) optimizing sensitivity. For each optimization criteria and metric, the significant differences (p < 0.05) for each input dataset are marked with 10th–90th percentiles of EHG parameters + obstetric input data; 50th percentile of EHG + obstetric input data; 10th–90th percentiles of EHG parameters; 50th percentile of EHG parameters. Significant differences between the two optimization criteria for the same input data set are marked with *.

Figure 4

Mean values of different KNN classifier metrics for validation datasets in the 30 data partitions: (a) optimizing F1-score (b) optimizing sensitivity. For each optimization criteria and metric, the significant differences (p < 0.05) for each input dataset are marked with 10th–90th percentiles of EHG parameters + obstetric input data; 50th percentile of EHG + obstetric input data; 10th–90th percentiles of EHG parameters; 50th percentile of EHG parameters. Significant differences between the two optimization criteria for the same input dataset are marked with *.

Figure 5

Mean values of different classifier metrics for validation datasets in the 30 data partitions obtained for the best RF, ELM and KNN classifiers. Significant differences (p < 0.05) of the classifiers and metrics with the others are marked with RF_{F1_2}; ELM_{F1_2}; KNN_{F1_2}.

Figure 6

Average receiver operating curves (ROCs) for training, validation and test datasets for the ELM_{F1_2}.