Semester of Graduation
Fall 2025
Degree Type
Dissertation/Thesis
Degree Name
Masters in Artificial Intelligence
Department
College of Computing and Software Engineering
Committee Chair/First Advisor
Chen Zhao
Second Advisor
Michail Alexiou
Third Advisor
Manohar Raavi
Abstract
In the realm of medical imaging, pipelining non-linear clinical data plays a critical role in aiding doctors in making diagnoses, especially for cardiologists. There exists no known way to digitize heart data for artificial intelligence training, the intricate coronary artery data at a given timestep in the heart cycle, a detailed 3D representation of the artery, nor the appropriate physical metrics to go along with the visualization. This study briefly describes theoretical and practical methods of processing coronary artery image data and performing regression by proposing a novel diffusion model, inverted conditional diffusion, to predict the blood pressure at given points. The methods are built on top of coronary computed tomography angiography and pipeline computational fluid dynamics. The impact of this study will enable cardiologists to diagnose coronary artery diseases such as stenosis programmatically, with ease, speed, and efficiency. A comprehensive comparative analysis was undertaken to evaluate the performance of the data pipeline with several deep learning models. The results showcased a consistent yet flexible pipeline with an average R-squared of 64.42% on multiple arteries. The study also provided a practical guide to set up the system using modern open-source tools. The study sets the stage for future enhancements in non-invasive coronary artery disease detection, blood pressure prediction, and non-linear regression.
Comments
We acknowledge the start-up funding for Dr. Chen Zhao provided by Kennesaw State University. This research was supported by an Interdisciplinary Seed Grant from Kennesaw State University (Grant Number 000149) and an AIREA grant from the American Heart Association (grant number 25AIREA1377168).