Degel Agoreyo
BME PhD Proposal Presentation

Date: 2026-08-26
Time: 10:00am
Location / Meeting Link: https://gatech.zoom.us/j/98134044735?pwd=HvPSTF5tXbUaDGHM3of7KAWYuEERlW.1

Committee Members:
Rudolph Gleason, PhD (Advisor) Maribeth Coleman, PhD Cassie Mitchell, PhD Lakshmi Dasi, PhD Alexander Adams, PhD


Title: A longitudinal, affordable machine learning informed web application for assessing cephalopelvic disproportion in Ethiopia

Abstract:
Although the number of deaths from preventable pregnancy-related complications daily has reduced from 800 in 2020 to 700 in 2023, a maternal death is still recorded every two minutes. Sub-Saharan Africa alone accounts for 70% of maternal deaths worldwide. In Ethiopia, where malnutrition, short stature, and underage pregnancies are common, obstructed labor is an important contributor to pregnancy complications, with a large proportion attributable to cephalopelvic disproportion (CPD). CPD is a type of labor obstruction caused by a mismatch between the fetal head and the mother's pelvis. Complications include fetal asphyxia, sepsis, stillbirth, uterine rupture, obstetric fistula, and hemorrhage, which is one of the leading causes of maternal mortality. These complications are preventable, but a lack of technical expertise, delays in referrals to specialized hospitals, and limited access to specialized facilities increase the risk of adverse outcomes. Reports show that women referred to specialized hospitals arrive more than 12 hours after labor has begun and often have to travel over 10 kilometers to reach the facility. Pregnant women are also reported to miss one or more antenatal visits. This creates a need for a longitudinal predictive CPD risk-stratification tool for early referral of high-risk mothers to facilities where they can receive appropriate obstetric care for safe delivery. Previously, in a cross-sectional study, we explored a combination of maternal body measurements from tape measurements and 3D point clouds at ≥36 weeks of gestation as predictors of CPD models, which showed great promise. In this proposal, we expand our work using a growth approach that incorporates maternal body measurements across a broader range of gestational ages (≥12 weeks), including women who may miss one or more antenatal visits. We also increase the sample size to ensure adequate sample power and generalizability of the risk stratification model, including women from different regions. This proposal will develop a longitudinal CPD risk stratification model that leverages tree-based, linear, kernel-based, and neural network approaches, using maternal tape measurements. It will also use 3D point clouds to better understand potential variability in tape measurements to improve model performance. The developed model will then be integrated into a web application and clinically evaluated to assess ease of use, acceptability, accuracy, and to understand potential barriers users may encounter. The expected outcome of this work is a longitudinal risk-stratification web application that predicts CPD risk, is easy to use, and is accepted by users in rural communities and healthcare settings in Ethiopia.