Automatic Evaluation of Fetal Heart Rate Based on Deep Learning

Abstract

Fetal heart rate (FHR) monitoring has been widely applied to assess the status of fetus during pregnancy and labor in clinical practice. However the traditional way to analyze FHR highly depends on doctors’ experience, and sometimes wrong judgments can lead to unnecessary actions such as cesarean section. Thus automatic analysis of FHR in electronic fetal monitoring (EFM) through computer has been constantly tried and studied. In this work, we propose a convolutional neural network (CNN) model based on a weighted voting mechanism to divide the FHR as normal or pathological state. In the meantime, the multi-model training method based on down-sampling algorithm is used to deal with imbalanced data. In order to evaluate the effectiveness of the proposed CNN combined with the multi-model training method, we test and analyze it on an open database named CTU-UHB. The experiment results show that our method performs well and stable on this dataset.