Neural Network Approach to Estimate Fetal Weight in the Late Third Trimester of Pregnancy - Project Summary and Abstract Fetal weight estimation, or the assessment of antenatal fetal weight for the purposes of growth tracking and labor planning, is a critical component of safe prenatal care. Estimations currently rely on ultrasound-derived measurements of specific fetal planes to indirectly assess growth and wellbeing. The standard fetal biometric measurements for the estimation of fetal weight (biparietal diameter, head circumference, abdominal circumference and femur length) are poorly correlated to actual fetal weight, defined as the measurement of newborn weight in grams at birth. For newborns who are above 4,000 grams at birth, current error estimates of fetal weight in the late-third trimester of pregnancy are only accurate approximately 40% of the time. By no longer relying on fetal biometric measurements, data science approaches have the potential to estimate fetal weight with lower bias and errors compared to standard regression methods. To date, no studies have used ultrasound images, not just the fetal measurements, as input into a neural network approach to estimate fetal weight. The overarching goal of this proposal is to develop the skills and training necessary to lead the advancement of data science for use in clinical assessment during pregnancy. Using existing ultrasound imaging and birth certificate data (n=17,478 patients) from the University of Rochester (UR) Medicine Hospitals and the Finger Lakes Regional Perinatal/Obstetrics Data System (PDS), and n= 310 patients in the R01 study, Understanding Pregnancy Signals and Infant Development (UPSIDE: R01HD083369), the specific aims are: 1) To determine the maternal (i.e., body mass index) and fetal factors (i.e., growth measurements) that increase the discordance between the estimation of fetal weight by the Hadlock formula and actual birth weight of neonates using birth certificate data from the PDS, 2) To evaluate the accuracy of a CNN algorithm on ultrasound images in the third trimester to estimate fetal weight compared to the Hadlock formula, and 3) To test the effectiveness CNN algorithm on new ultrasound images from the UPSIDE study. This proposal will leverage the expertise of Dr. Caitlin Dreisbach’s mentorship team, computational resources, and the exceptional research environment at the UR School of Nursing, Goergen Institute for Data Science, and the Rochester Institute of Technology. Results from this study have the potential to change practice and improve clinical assessments during the late third trimester of pregnancy. The research study and mentored training included in this award allows Dr. Dreisbach to establish her long-term career goal of becoming an independent investigator with expertise in the translation of data science to obstetric clinical care.
