Jieqiong Wang COBRE Abstract

Project Summary/Abstract: Wang Supplement Project

Effective prenatal screening for congenital heart disease (CHD) is critical for safeguarding maternal and infant health. However, rural healthcare systems, particularly in Nebraska and the broader Midwest, face significant challenges in providing accurate CHD diagnosis due to limited access to trained ultrasound operators and unequal distribution of medical resources. These disparities result in poorer health outcomes for women in rural areas. Existing artificial intelligence (AI) and machine learning (ML) approaches hold promise for improving CHD diagnosis but are hindered by lower-quality imaging, an inability to identify CHD subtypes, and a lack of attention to health disparities, particularly for minority and rural populations. To address these issues, we propose developing an AI/ML model capable of accurately diagnosing specific CHD subtypes, leveraging diverse data from both rural and urban settings. Additionally, our framework includes a novel transfer learning approach to directly address health disparities and enhance access to high-quality CHD screening for women in rural areas.

 We hypothesize that this CHD subtype diagnosis system, built on a diverse dataset from both rural and urban environments, will enhance diagnostic accuracy and mitigate health disparities in rural populations. Under our parent award (P20GM152326), in this supplemental project, we propose to develop novel AI/ML models for prenatal screening of CHD subtypes and reducing rural health disparity. To achieve our goal, we will undertake two specific aims.

 Aim 1, to develop a novel AI/ML model for the prenatal diagnosis of CHD subtypes. By using both rural and urban data, we will create a model that classifies seven guideline-recommended cardiac views from fetal ultrasound images, including standard screening scans and echocardiograms. This model will differentiate between critical and non-critical CHD subtypes, enabling timely and precise interventions that improve maternal health outcomes, particularly in underserved communities.

 Aim 2, to reduce health disparities in prenatal CHD diagnosis in rural areas through a transfer learning framework. We will evaluate how factors such as geographic location and race affect ML model performance. Our strategy includes balanced training data, subgroup-specific models, and a transfer learning approach that uses data augmentation to fine-tune models for underrepresented groups, ensuring both fairness and diagnostic accuracy across diverse populations.

 Our multidisciplinary team, led by experts in machine learning, maternal-fetal medicine, and biostatistics, is uniquely equipped to address these critical challenges. Successful completion of this project will enhance prenatal CHD care, improving early diagnosis and enabling timely interventions for women in rural and underserved communities. By focusing on reducing health disparities and promoting equitable access to advanced diagnostic tools, our work will pave the way for improved maternal and infant health outcomes across the Midwest and beyond.