Uterine contraction (UC) signals record the pressure of uterine contractions in pregnant women and are used, in conjunction with a physician’s internal examination, to diagnose the stage of labor progression. If the stage of labor could be automatically estimated from UC signals, it would reduce the number of internal examinations by physicians, thereby alleviating the burden on pregnant women, lowering the risk of infection associated with such examinations, and helping address the shortage of medical personnel.

This paper proposes UCscan, an analytical method that automatically estimates the latent stages of labor progression from UC data measured during childbirth. UCscan extracts contraction waveforms and their occurrence intervals from UC signals and aggregates waveform patterns common across sequences into interpretable components, thereby estimating the irreversibly progressing stages of labor. Experiments using real-world data demonstrate that the proposed method achieves higher estimation accuracy and significantly improved computational efficiency compared to existing methods.

Since UC signals are difficult to obtain, research on their modeling and analysis has not progressed substantially, making the very act of addressing this problem highly valuable. In addition, large-scale datasets are not available, which makes it difficult to apply approaches based on deep learning or foundation models. In contrast, this study effectively incorporates multiple medical insights and carefully models waveform patterns, referred to as components, along with irreversible stage transitions based on a hidden Markov model (HMM), thereby proposing a new model, the UCscan. This method is evaluated using actual UC data collected at university hospitals and similar institutions, demonstrating improved stage assignment accuracy and significantly reduced computation time compared to existing approaches such as TS2Vec and conventional HMM-based methods. Furthermore, this study includes applied experiments addressing the issue of missing values in real UC signals, resulting in an experimental design that considers practical applications.

This paper not only has practical value through the real-data analysis of UC signals but also presents a highly interesting approach from the perspective of the mathematical modeling of time-series signals. It makes a significant contribution to the fields of biosignal analysis and time-series modeling and can be highly regarded as a paper worthy of a best paper award.