اميررضا اسدي

Fetal Electrocardiogram (FECG) provides helpful information regarding the fetus's health and heart condition during pregnancy. Since the FECG is extracted from the maternal abdominal electrocardiogram (AECG) signal, this task can be considered a source separation problem. This study presents two methods of extracting FECG from single-channel AECG signals based on independent component analysis (ICA) and deep learning. The proposed algorithms utilize the method of time delay to map one-dimensional time series to multi-dimensional series, then in order to extract FECG, multi-dimensional series are regarded as input of ICA or convolutional neural network (CNN). In the ICA-based method, after the reconstruction of the time series, the components are classified into two groups, fetal and mother, using power spectral analysis. In the method based on deep learning, the network is a modified version of the U-Net architecture, where the network's input is the matrix, and the output is the FECG signal. Using a two-dimensional network (2D), this model obtains 2D features of AECG and improves the FECG extraction accuracy. We eva‎luated the performance of the proposed methods on the Set-A of 2013 PhysioNet/Computing in Cardiology Challenge database (PCDB) and the abdominal FECG database (ADFECGDB) utilizing fetal QRS complex (FQRS) analysis. F1 scores in the ICA-based method on PCDB and ADFECGDB are 95.76% and 95.99%, respectively, and in the deep learning-based method, 97.97% and 96.14% respectively. Results of the proposed methods have been compared with that of other state-of-the-art deep learning models such as Generative adversarial Network (GAN), Residual Convolutional Encoder-Decoder Network (RCED-Net), U-Net, and W-Net, which demonstrate the effectiveness of the proposed methods. Additionally, the statistical measures exhibit that the proposed deep learning model on the PCDB database is superior to other single-channel methods. Therefore, the proposed methods are suitable for long-term monitoring of the mother and fetus.

سيگنال الكتروكارديوگرام جنين , شبكه‌ي U-Net , تحليل مؤلفه‌ي مستقل , روش تأخير زماني , سيگنال الكتروكارديوگرام شكمي

Fetal Electrocardiogram Signal , convolutional neural network , independent component analysis , method of time delay , Abdominal Electrocardiogram signal

Amirreza Asadi

Dr. Aboozar Ghafari