Abstract

Segmenting and reconstructing cardiac anatomical structures from magnetic resonance (MR) images is essential for the quantitative measurement and automatic diagnosis of cardiovascular diseases [1]. However, manual evaluation of the time-series cardiac MRI (CMRI) obtained during routine clinical care are laborious, inefficient, and tends to produce biased and non-reproducible results [2]. This thesis proposes an end-to-end pipeline for automatically segmenting short-axis (SAX) CMRI images and generating high-quality 2D and 3D meshes suitable for finite element analysis. The main advantage of our approach is that it can not only work as a stand-alone pipeline for the automatic CMR image segmentation and mesh generation but also functions effectively as a post-processing tool for improving the outcomes of deep learning methods. Our results indicate that the segmentation accuracy outperformed the traditional U-Net-based approach by as much as 82.5% (percent increase in Dice score) for 5 patient types. The mesh models generated from our contoured segmentations had minimized mean distance error of less than 1.3 pixels and optimized mesh quality with an average Kupp index greater than 0.8.

Committee Chair

Jon Silva

Committee Members

Jonathan Moreno Christian Zemlin

Comments

Revised according to comments

Degree

Master of Science (MS)

Author's Department

Biomedical Engineering

Author's School

McKelvey School of Engineering

Document Type

Thesis

Date of Award

Summer 9-8-2023

Language

English (en)

Author's ORCID

https://orcid.org/0000-0003-3546-6065

Download

Share

COinS