Supervised Convolutional Encoder-Decoder with Gated Linear Units for Detecting Fetal R-Peaks

Noninvasive fetal electrocardiography (ECG) is prevalently used for monitoring fetal heartbeats during pregnancy due to its affordability, ease of use, and constant monitoring capability. A crucial aspect of noninvasive fetal ECG is detecting the R-peak series from the abdominal electrode signal, which is a fundamental baseline for determining fetal heart rate. This paper explores the direct detection of R-peaks by assigning categorical labels to each member of the observed values in the ECG sequence and proposes a convolutional encoder-decoder network and training strategy for processing the sequence annotation task. Specifically, the encoder is a stacked convolutional layer equipped with a gating linear unit (GLU), and the decoder is a recurrent neural network. The GLU convolutional layer can effectively extract and aggregate the features to improve the generalization ability. To address the issue of unbalanced sequence labels, we adopt and fine-tune the focal loss function, promoting superior prediction and faster convergence. The experimental results suggest that the proposed method can achieve promising performance on two benchmark datasets. The versatility of our approach is validated through tests of different label encoding strategies, demonstrating its potential for other complex fetal ECG labeling tasks.