Original title: Empirical mode decomposition and interpretable machine learning for preterm birth classification from electrohysterography
Umesha Tilakarathna, Senith Jayakody, Kalana Jayasooriya, Roshan Godaliyadda +3 · August 2026
Predicting preterm birth safely and early is hard. Doctors can measure electrical signals from the uterus through the mother's abdomen, but standard methods sometimes produce inflated results during testing. Researchers tested a mathematical technique that breaks complex electrical signals into simpler underlying components called intrinsic mode functions. They tested these components on pregnancy records to distinguish preterm from full-term births. The first component proved to be the most accurate for classification. It separated the groups better than traditional filtered time signals. The researchers found that preterm signals showed smaller, more evenly spaced peaks, less temporal energy, and higher entropyA measure of randomness or unpredictability in a signal., meaning the signals were more disordered. These patterns help models identify pregnancy risks more reliably.
Imagine trying to tell if a car engine is about to fail just by listening to its overall hum. The sound is a mix of many vibrations, and subtle clues are buried in the noise. Simply recording the whole sound and analyzing it all at once might miss the early warning signs. In this paper, the researchers break the uterine electrical signal into its basic rhythmic components, and find that the fastest component—the first one—carries the clearest information about whether the pregnancy will end early. By focusing on that component, they can classify preterm risk more accurately.
Preterm birth is a leading cause of infant death and lifelong health problems, yet current risk assessment is unreliable. This study shows that analyzing the electrical activity of the uterus with a specific signal decomposition technique can distinguish preterm from term pregnancies with about 83% accuracy. If validated in larger studies, this could become a simple, non-invasive screening tool to identify at-risk pregnancies earlier.
The study uses only 26 recordings from a single public dataset (TPEHGT), which is too small to draw firm conclusions. The results may not generalize to other populations or recording setups. The analysis is limited to IMF1 and fixed 3-minute windows; other decomposition methods or window sizes could yield different results. The paper explicitly calls for validation in larger independent cohorts before clinical use.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Empirical mode decomposition and interpretable machine learning for preterm birth classification from electrohysterography — Umesha Tilakarathna, Senith Jayakody, Kalana Jayasooriya et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.