TR-index: Semantic Characterization for Non-invasive Fetal ECG Signal Quality Assessment

Abstract

The non-invasive fetal electrocardiography (NI-FECG) plays an important role in prenatal monitoring. However, the NI-FECG is inevitably affected by a variety of non-stationary interferences, making the detecting of fetal QRS (FQRS) complexes a challenging task. In this study, we propose a novel framework that dynamically assesses the quality of NI-FECG to improve multichannel FQRS detection. The novelty of the framework relies on assessing the signal quality in semantic level. Specifically, each of the detected FQRS can be considered as a semantic unit, which contains the basically semantic information (true R peak (TR) or false R peak (FR)). By obtaining the possible condition of correct detections, the signal quality of each channel can be estimated. Further, the performance of the corresponding channel on the FQRS detection becomes assessable. The procedure contains four stages: Firstly, single-channel FQRS detection is employed on each channel individually. Secondly, a convolutional neural network (CNN) is used to classify the detected FQRS as either TR or FR. Thirdly, the TR-index is used to quantity the signal quality of single channel. And the channel with the highest TR-index is selected as the optimal channel. Finally, a robust strategy that mimics the human approach is used to fuse the TRs of all channels into a new channel. Results on two databases show that the proposed method is effective in FQRS detection achieving 97.81%Precision, 98.29%Recall and 98.05%$F_{1}$. This work sheds some light to the topic of NI-FECG signal quality assessment.