During the last 20 years, in the context of child and maternal mortality rate reduction, several medical studies aimed at increasing current digital monitoring techniques during delivery, with the purpose of reducing risks connected with prolonged and arrested labor. Several strategies for minimizing the obstetric risk factors for failure of labor to progress have been tested during the years. Among different alternative monitoring techniques, uterine electromyogram (EMG) also called electrohysterogram (EHG), has shown to produce some of the most promising biophysical markers for both labor arrest and prolonged labor. Trans-abdominal EMG is a noninvasive technique, which enables measurement of electrical activity produced by uterine muscles through the maternal abdominal surface.
The purpose of this study is to analyse EMG activity generated by uterine muscular fibers by designing, building and testing a software able to process raw signals acquired during labor. Given the fact that previous studies revealed a correlation between power density spectrum of contractions and labor progress, we planned to develop a software capable to extract and isolate contractions in each signal and to calculate the corresponding EMG power spectral density peak frequencies, in order to use these information to assess the risk of labor arrest. The ultimate goal of this research is to determine whether this technique could be a valid identifier of inefficient contractions during labor. It was conducted in cooperation with “Ospedale dei bambini Vittore Buzzi” of Milan, whose team provided us with EMG signals acquired during the period 2010-2013 from nulliparous women.
A specific algorithm was designed and developed. The algorithm, designed with a structured programming paradigm, consists of 5 modules, each fulfilling a macroscopical functional data processing.
In an iterative process, each modules acquires data made available from its ancestor section, refines data with its own processing, and produces output to the following module.
Module1: Acquisition of RAW electromyograph data.
Module2: Signal noise filtering of acquired raw data.
Module3: Contraction detection on denoised signals.
Module4: Evaluation of contractions detected and characteristics computation. Module5: labor information exportation.
The final output is a set of detailed information about each contraction in each labor, stored in a specific data structure, in accordance with a provided recordset. The contraction rate computed by the algorithm was compared to the output produced by commercially available software, revealing an excellent ability to identify contractions and confirming many of the expectations defined during the design- analysis phase.
Thanks to its efficient signal processing libraries, the environment chosen for the first level implementation was Matlab©. A simple and specific user interface was built in order to allow a more intuitive testing and results collection, to simplify integration for future development and to grant a good user-experience during medical-side usage of the software.
A first statistical analysis has been performed on the output produced by the algorithm. It revealed a promising correlation between mean and median PDS peak frequencies and labor progression status. In particular, we observer a median PDS peak value in labors leading to CD significantly lower (0,4463 Hz) than in women delivering vaginally with or without usage of delivery rooms devices (0,4669 Hz), (P <0.05, Mann–Whitney double tail rank-sum test). Results have been confirmed by further analyses on each first and second stage labor phase (L1, L2, L3, Push). Starting from the results of the work, the medical team at “OSPEDALE dei BAMBINI VITTORE BUZZI” will soon be undergoing clinical investigation on the interaction between contraction features and first and second stage labor arrest.
