Arye Nehorai, Eugene and Martha Lohman Professor of Electrical Engineering in Washington University’s Preston M. Green Department of Electrical and Systems Engineering (ESE) in St. Louis, and Uri Goldsztejn, a graduate student in the Department of Biomedical Engineering who works under his direction, trained Deep learning electrical data (HEG) model to predict preterm birth as early as 31 weeks gestation. The results of their research were published in PLoS One on May 11th.
Almost one in ten babies in the world is born prematurely, i.e. before the eighth month of pregnancy, which can lead to permanent neurological deficits and is one of the main causes of infant mortality. In France, there are almost 55,000 preterm births each year, with 15% of very preterm babies (born between 6 and 7 months of pregnancy) and 5% of very preterm babies being born even earlier.
Prevention of preterm birth is a public health problem. His prediction would make it possible to set up aftercare and medical care aimed at delaying birth.
Professor Arye Nehorai explains:
“Our method predicts preterm delivery using electrohysterogram measurements and clinical information collected at approximately 31 weeks gestation, and has performance comparable to clinical standards for detecting imminent labor in women with symptoms of preterm labor.”
This research, which developed the first method to predict preterm birth as early as 31 weeks using EHG measurements that achieve clinically useful accuracy, builds on earlier work from Arye Nehorai’s lab. In this study, Arye Nehorai and his collaborators had developed a method to estimate the electrical current in the uterus during contractions using magnetomyography, a non-invasive technique that maps muscle activity by recording abdominal magnetic fields generated by currents. electricity in the muscles.
It also builds on research by Arye Nehorai and Uri Goldsztejn, recently published in Biomedical Signal Processing and Control, which describes a method of statistical signal processing to separate the electrical activity of the uterus from basic electrical activity, such as that of the female heart , in multidimensional EHG measurements to identify separate uterine contractions more accurately
EHG measurements and clinical data
The EHG, electrohysterogram or uterine electromyogram, allows the electrical activity of the uterus to be recorded using a device consisting of electrodes placed on the abdomen, connected to an amplifier of electrical signals and connected via WiFi to software for signal analysis.
For their study, the researchers therefore used EHG measurements and clinical information from two public databases, such as age, gestational age, weight and bleeding in the first or second trimester.
They trained a deep learning model using 30-minute EHG data taken on 159 women who were at least 26 weeks pregnant. Some recordings were made during regular check-ups, while others were recorded from hospitalized mothers with symptoms of preterm labour. Of these women, almost 19% gave birth prematurely.
Uri Goldsztejn says:
“We predicted pregnancy outcomes from EHG recordings using a deep neural network because neural networks automatically learn the most informative features from the data. The deep learning algorithm performed better than other methods and provided a good way to combine EHG data with clinical information.”
They also showed that predictions could be made based on shorter EHG records, even less than five minutes, without significantly affecting the accuracy of the predictions.
The two researchers now want to build a device to implement their method and collect data from a larger cohort of pregnant women to improve their model and validate the results.
Item references:
McKelvey School of Engineering blog, Beth Miller
Goldsztejn U, Nehorai A. Predicting prematurity from electrohysterogram recordings using deep learning. PLoS One, May 11, 2023. DOI: https://doi.org/10.1371/journal.pone.0285219
.