Eric Macon
Research Mentor: Domingo Uceda
Mentor Department: Department of Surgery, Medicine
Author(s): Eric Macon, Domingo Uceda
Session: Session 7 (4:00 PM – 4:50 PM)
Presentation Type: Poster 99
Abstract
The continuous functionality of the myocardium, muscular tissue of the heart, is dependent on through consistent circulation of Oxygen. Certain dysfunctions impact this circulation, like myocardial ischemia by constricting coronary passageways which can lead to heart attacks. Building a stronger understanding of the hemodynamics and transport dynamics of this system requires a more intimate knowledge of the coronary architecture than the long used 2D histological data we have today. In its place, more accurate and advanced 3D reconstruction of the coronary network is certainly possible through the use of modern imaging and automated processing techniques. Our research utilizes fluorescent microscopy and tissue-optimized optical clearing to produce cellular level detail with submicron resolution of volumetric images of heart tissue. To automate vascular segmentation, we applied a self-configuring deep learning framework (nnU-Net), which adapts preprocessing, network design, and training strategies to the dataset. This framework was first pre-trained with a cleared rat heart dataset, then developed with time consuming but accurate manual segmentation of swine heart datasets. Initial segmentation using a pre-trained deep learning model showed limited accuracy on our swine heart dataset. After exposing the pre-trained model to manually annotated swine myocardial tissue, segmentation performance quality improved significantly. This enhancement and others allowed our self-configuring framework to more accurately detect complex coronary vessel structures throughout the heart tissue, closely matching manual annotations. After development, our model is projected to offer both a scalable alternative to manual annotation, and preserve accuracy across complex vessel geometries. By reconstructing 3D vascular networks with high fidelity, we deepen our insight into myocardial perfusion patterns and the structural underpinnings of coronary dysfunction. Ultimately, this method enables broader, high-throughput studies of cardiac microcirculation. Supporting advancements in cardiovascular research, personalized medicine, and clinical diagnostics for conditions like ischemia and heart failure.
