Project Cross-Modality Deep Learning for Echocardiographic Assessment of Aortic Root Mechanics Program DUIRI - Discovery Undergraduate Interdisciplinary Research Internship Term Fall 2026 Status Accepted Research Area Global Health Description According to the World Health Organization, cardiovascular disease is the leading cause of death worldwide, with over 75% of deaths resulting from cardiovascular disease being in lower- and middle-income countries. One type of cardiovascular disease is the development of aortic aneurysms. Aortic aneurysms result from weakness in the aortic wall that leads to wall expansion and ballooning. Aortic wall dissection can follow this expansion, ultimately leading to rupture of the aorta, which has a very high mortality rate. This project aims to improve how we evaluate aortic disease by analyzing aortic root mechanics, such as how the vessel expands and relaxes during the cardiac cycle. While prior work has focused on transthoracic echocardiography (TTE), this study extends these methods to transesophageal echocardiography (TEE), which provides higher-resolution images but is more challenging to analyze automatically. We will test the hypothesis that deep learning models trained on TTE can be adapted to TEE using cross-modality transfer learning, enabling robust and automated extraction of biomechanical metrics from TEE data. Echocardiographic data from pediatric patients at Yale-New Haven Children’s Hospital will be used to evaluate model performance and to test the hypothesis, that will eventually affect the clinical diagnosis and monitoring of pediatric ATAAs, based on the American Heart Association (AHA) guidelines. Supervisor Charles A Bouman Mentor Shubh Parag Mehta Type of work required Students involved in this project will:
- Assist in adapting existing deep learning segmentation models trained on transthoracic echocardiograms to transesophageal echocardiography data using transfer learning techniques
- Perform preprocessing and standardization of echocardiographic cine loops across imaging modalities
- Evaluate model performance using metrics such as Dice coefficient, Intersection-over-Union (IoU), and boundary-based measures
- Extract quantitative biomechanical metrics, including diameter-time curves and strain-based measures, from model outputs
- Conduct statistical comparisons between automated measurements and clinician-derived annotations
- Contribute to exploratory analyses comparing biomechanical features across patient groups
Students will gain experience in medical image analysis, deep learning, and translational cardiovascular research. Websites https://engineering.purdue.edu/cvirl
https://pubmed.ncbi.nlm.nih.gov/39299353/
Paik, Joshua; Mehta, Shubh P.; Sivakumar, Samskrithi; Dinklage, Felix; Lee, Ahhyun; Landis, Benjamin J.; and Goergen, Craig J., "Improving Echocardiographic Aortic Aneurysm Assessment in Marfan Syndrome Patients" (2025). Discovery Undergraduate Interdisciplinary Research Internship. Paper 60.
https://docs.lib.purdue.edu/duri/60
Qualifications Students should be highly interested in medical imaging. Previous coding experience is required in Python and/or MATLAB. A basic understanding of cardiovascular anatomy will be helpful but not required. Credit Hours 0 Weekly Hours 10 (estimated)