Fetal monitoring will usher in major changes
Researchers in Iran have used a deep neural network (DNN) to extract a fetal electrocardiogram (ECG) from a single abdominal ECG channel. Their method, described in Physiological Measurements, may improve fetal monitoring in the future.
How to isolate fetal ECG?
Currently, the electrical activity of the fetal heart is measured by an electrocardiogram obtained from electrodes with ECG leads placed on the expectant mother's abdomen. Clinicians can use the fetal ECG to assess fetal health and diagnose abnormalities.
The challenge? It is difficult to separate fetal ECG signals from abdominal ECGs, which contain signals from the fetus (“fetal ECG”) and mother (“maternal ECG”), as well as signals from sources of interference such as muscle contractions . This task becomes more demanding towards the end of pregnancy, when the amplitude of the fetal ECG signal is comparable to that of the maternal ECG.
The study's lead author, Arash Rasti-Meymandi, a graduate student at Iran University of Science and Technology, and his colleagues came up with an approach that relies on DNNs to solve the problem.
Rasti-Meymandi was inspired by Unets, convolutional networks commonly used for medical image segmentation tasks. He and his collaborator Aboozar Ghaffari applied a modified version of Unet to first extract the ECG of the pregnant woman and then the ECG signal of the fetus.
"Unet outperforms other techniques in image segmentation," Rasti-Meymandi said. "To extract different components of the abdominal ECG, we detected abdominal ECG signals at different resolutions (similar to the process used in the Unet model) )."
The researchers' DNN, called AECG-DecompNet, uses two sub-networks in series to extract fetal ECGs from a single-channel abdominal ECG. The first sub-network extracts the maternal ECG; the second is the fetal ECG. The researchers separately trained the two sub-networks using simulated ECG signals, and then evaluated the sub-networks using simulated and real abdominal ECG recordings.
Using a graphics processor, the researchers' DNN can process four seconds of abdominal ECG recordings in about one second.
The future of DNN and fetal ECG
Unlike other signal noise reduction methods that require a reference ECG modality (P, Q, R, S, and T waveforms that indicate the heart's electrical activity), multi-channel ECGs, or both, the researchers' method requires only one channel, i.e. Can. This not only improves the mother's comfort during ECG acquisition, but also requires fewer resources and less time to implement compared to traditional ECG recording and signal extraction methods.
The researchers also found that their method better preserved the shape and structure of fetal ECG signals relative to other methods -- all five waveforms were well preserved, allowing detection and diagnosis of fetal abnormalities.
"The main result of this study is the effectiveness of using DNNs to efficiently extract fetal ECG signals from single-channel abdominal recordings," Rasti-Meymandi told Physical World. "We are currently working on more sophisticated algorithms ... to further Improve the accuracy of heart rate extraction."
The team is also working on ways to implement DNNs in real-time on smartphones.
Limitations of their approach include a potential over-reliance on the training dataset, especially with weak fetal ECG signals, and error propagation from the first sub-network to the second.