SEM-based study for interpretability of intelligent prenatal fetal monitoring models

EasyChair Preprint 1722

Abstract

Widely available computerized cardiotocography (CTG) data and machine learning methods present an opportunity to improve the accuracy and scalability of automated CTG analysis for prenatal fetal monitoring. However, their interpretability is not clear enough, due to superficially dependent on data. In this paper, we established a measurement model through the fetal prenatal monitoring clinical knowledge and then use the SEM model to derive the relationship between the variable category (baseline category, variability category, acceleration category, deceleration category, uterine contraction) in the measurement model and explore their impact on the fetus status. In particular, The learned models concluded that the variant category predicts the baseline category and that the contraction has a predictive effect on the deceleration category, validating the SisPorto 2.0 (CTG analysis program) and prenatal fetal monitoring clinical knowledge, respectively. In addition, the variability category and the acceleration category have a greater impact on the identification of fetal status, while the baseline category has less impact, explaining the importance distribution of CTG features in weighted random forests.

Keyphrases: Cardiotocography, Prenatal fetal monitoring, Random Forest, interpretability, structural equation model

@booklet{EasyChair:1722,
  author    = {Li-Ting Huang and Rui-Chu Cai and Zhi-Ying Jiang and Li Li and Qin-Qun Chen and Jia-Ming Hong and Zhi-Feng Hao and Hang Wei},
  title     = {SEM-based study for interpretability of intelligent prenatal fetal monitoring models},
  howpublished = {EasyChair Preprint 1722},
  year      = {EasyChair, 2019}}