Interpreting Uterine Contractions Using Physiological Models and Statistical Signal Processing

Pregnancy and labor are important stages in every mother and child’s life. Unfortunately, complications during pregnancy and labor are common. Every day, globally, about 800 mothers and 7000 newborns in their first month of life die from medical complications. Additionally, these complications result in about 5000 stillbirths daily around the world. Preterm births, labor dystocia, and postpartum hemorrhage are leading causes of maternal and child morbidity and mortality and are associated with inappropriate uterine contractile activity. Preterm births, i.e., those that occur before completing 37 weeks of gestation, are a leading cause of child morbidity and mortality. Preterm birth are a major health burden since about 10% of all births are preterm. Labor dystocia refers to labors that progress slowly. Labors may progress slowly because the uterine contractions are too weak to deliver the child. Slowly progressing labors are associated with various medical complications including infections, neonatal distress and asphyxia, and uterine rupture. Postpartum hemorrhage results from insufficient uterine contractions after labor and is the main cause of maternal mortality in low-income countries. Various technologies and treatments have been developed to predict uterine contraction disorders and to improve their outcomes. These technologies include devices for monitoring uterine contractions and biochemical tests that infer uterine conditions. The drugs used to treat these conditions usually regulate uterine activity by modulating the ion currents responsible for uterine contractions. However, these technologies and treatments have limited efficacy. Moreover, these treatments can have serious side effects and cause additional medical complications. Two important limitations on developing safer and more efficient technologies and therapies for these conditions are our incomplete understanding of uterine electrophysiology and the difficulty of measuring uterine activity accurately. Here, we developed a uterine muscle fiber model and a statistical tensor decomposition method to address these limitations. Our fiber model incorporates various models of uterine cellular functions and a novel set of differential equations that simulates how the electrical and mechanical activities propagate along a uterine fiber. Using this model, we investigated how cellular excitability and intercellular coupling regulate electrical conduction and contractile force generation in uterine fibers. Among other observations, we found that cellular coupling plays a major role in regulating contractile force generation. We observed that if the cells are well coupled, then the fiber can generate a stable tension. However, the fiber may not generate significant force when cellular coupling is low. We developed a statistical tensor decomposition method to estimate uterine activity more reliably. Electrohysterogram (EHG) recordings measure uterine electrical activity noninvasively using abdominal electrodes. EHG measurements are compelling because they capture informative physiological activity and can be used to develop portable applications for monitoring uterine activity. However, localized electrical activity in the uterus generates distributed electric potentials at the abdomen. Furthermore, EHG measurements also record electric potentials that originate from other sources besides the uterine section directly below the electrodes. Our method estimates the localized uterine electrical activity reliably based on a statistical generative model. We assessed the performance of our method using simulated and real EHG measurements. Then, using our method and two public EHG databases, we found that the fraction of the myometrium that is recruited during contractions is a potential biomarker to monitor uterine contractions. We found that the fraction of the myometrium recruited during contractions is higher in pregnant mothers who eventually delivered preterm than in those who delivered at term. Moreover, we show that this metric increases towards labor and that it may be used to monitor labor contractions. Our results at the fiber level and at the whole organ level suggest that uterine coupling is an important characteristic of uterine contractions that can be used for monitoring and treating abnormal contractions. Our work advances our understanding of uterine electrophysiology and contributes to the development of better technologies and more efficient drugs to monitor and regulate uterine contractions. Lastly, we develop a machine learning model to predict preterm births based on clinical information and EHG measurements. Although imminent preterm birth can be predicted about one week in advance in mothers with symptoms of preterm labor, predicting preterm birth more than one week in advance in asymptomatic mothers remains elusive. EHG measurements have been proposed to identify pregnant mothers at high risk of preterm labor. However, the predictive accuracy obtained using EHG measurements is limited. Here, we developed a machine learning model that predicted preterm birth directly from EHG measurements more accurately than existing models. Our model and results advance the research efforts for developing a screening tool to identify pregnant mothers at high risk of preterm birth, facilitating targeted treatments to reduce the incidence of preterm birth and improve the outcomes