huomarc / mlfetalhealth

Due to the intricacies of the body of a fetus, as well as the rapid rate at which fetuses develop, fetal health care is one of the most difficult medical fields to practice effectively. Due to the challenges around fetal health care, reducing child and maternal mortality rates is of paramount importance to every modern country. The reduction of child mortality is reflected in several of the United Nations' Sustainable Development Goals. The UN expects that by 2030, member countries will effectively end preventable deaths of newborns and children under 5 years of age (Child Survival and the SDGs, 2021). All member countries are aiming to reduce neonatal mortality to at least as low as 12 deaths per 1,000 live births and under-5 mortality to at least as low as 25 deaths per 1,000 live births (Child Survival and the SDGs, 2021). Maternal mortality is intrinsically tied with child mortality and has continued to have wide racial/ethnic gaps, even today. Black, American Indian, and Alaska Native (AI/AN) women are two to three times more likely to die from pregnancy-related causes than white women, and that disparity only increases with age (National Center for Health Statistics). Thus, the necessity of mitigating child mortality as much as possible cannot be overstated. Electronic fetal monitoring or cardiotocography is a “visual representation” of uterine contractions and fetal heart rate, which has been recognized as a prominent indicator of fetal health since the 19th century (Petker, 2018). Certain fetal heart rate patterns are linked to non-reassuring fetal status, and therefore, fetal heart rate monitoring can help prevent poor fetal outcomes (Petker, 2018). The fetal heart rate monitor identifies the normal baseline rate and tracks variability, accelerations, and decelerations to provide insight into a baby’s level of stress, oxygenation, acidemia (increase in hydrogen ion blood concentration), and other vital signs (Petker, 2018). A host of models were applied, including XGBoost, LightGBM, Gradient Boosting, Random Forest, Multi-layer Perceptron, Support Vector, Decision Tree, K-Nearest Neighbors, Linear Support Vector, and Logistic Regression, to the cardiotocography data to predict fetal health outcomes and level of care.

Machine Learning Multiclass Classification of Fetal Health using Cardiotocogram Data

Due to the intricacies of the body of a fetus, as well as the rapid rate at which fetuses develop, fetal health care is one of the most difficult medical fields to practice effectively. Due to the challenges around fetal health care, reducing child and maternal mortality rates is of paramount importance to every modern country. The reduction of child mortality is reflected in several of the United Nations' Sustainable Development Goals. The UN expects that by 2030, member countries will effectively end preventable deaths of newborns and children under 5 years of age (Child Survival and the SDGs, 2021). All member countries are aiming to reduce neonatal mortality to at least as low as 12 deaths per 1,000 live births and under-5 mortality to at least as low as 25 deaths per 1,000 live births (Child Survival and the SDGs, 2021). Maternal mortality is intrinsically tied with child mortality and has continued to have wide racial/ethnic gaps, even today. Black, American Indian, and Alaska Native (AI/AN) women are two to three times more likely to die from pregnancy-related causes than white women, and that disparity only increases with age (National Center for Health Statistics). Thus, the necessity of mitigating child mortality as much as possible cannot be overstated. Electronic fetal monitoring or cardiotocography is a “visual representation” of uterine contractions and fetal heart rate, which has been recognized as a prominent indicator of fetal health since the 19th century (Petker, 2018). Certain fetal heart rate patterns are linked to non-reassuring fetal status, and therefore, fetal heart rate monitoring can help prevent poor fetal outcomes (Petker, 2018). The fetal heart rate monitor identifies the normal baseline rate and tracks variability, accelerations, and decelerations to provide insight into a baby’s level of stress, oxygenation, acidemia (increase in hydrogen ion blood concentration), and other vital signs (Petker, 2018). A host of models were applied, including XGBoost, LightGBM, Gradient Boosting, Random Forest, Multi-layer Perceptron, Support Vector, Decision Tree, K-Nearest Neighbors, Linear Support Vector, and Logistic Regression, to the cardiotocography data to predict fetal health outcomes and level of care.