Abstract
Fetal heart rate (FHR) recorded by cardiotocography (CTG) is crucial for assessing fetal well-being; however, it is occasionally subject to the issue of maternal heart rate being recorded. Therefore, the aim of this study was to develop a convolutional neural network model capable of classifying fetal-derived signals and non- fetal-derived signals from Doppler ultrasound (DUS) signals with an accuracy comparable to that of midwives, and to evaluate its effectiveness. The data consisted of DUS signals and CTG recordings collected from 425 cases in the obstetrics ward of a single facility. This study used 526 fetal-derived signals and 114 non-fetal-derived signals from the 425 cases, annotated by midwives. A one-dimensional convolutional neural network (1D-CNN) model was developed, and its classification performance was evaluated using the area under the curve (AUC). As a result, the model achieved an AUC of 0.928. In addition, the classification performance of the 1D-CNN was found to be within the range comparable to that of midwives. Therefore, this technology is expected to be applicable as a support tool for both CTG sensor placement and the interpretation of DUS signals.
