Abstract

Cardiomyopathy is a disease of the heart muscle that can change cell structure, protein organization, and overall heart function. Some forms of cardiomyopathy are genetically inherited, including Hypertrophic Cardiomyopathy (HCM), the most common inherited cardiomyopathy affecting nearly 1 in 200 individuals1,2, and Desmin-Related Cardiomyopathy (DRM), a rarer disease that affects approximately 1 in 2000 individuals3. Changes in cardiomyocytes’ cellular structure define cardiomyopathies, and these changes are typically studied with fluorescence microscopy. However, manual image analysis is slow, labor-intensive, and often gives inconsistent results, especially when images are analyzed by non-experts. This thesis presents an automated computational framework for analyzing cardiomyocyte structure in images obtained from both two-dimensional (2D) and three-dimensional (3D) datasets. A key innovation of my work is the ability to identify (segment) individual cells in highly dense culture

environments with abundant cell-cell contacts – a setting in which unambiguous segmentation of individual cells can be challenging.

To overcome this fundamental challenge, I created a computational workflow that uses deep learning-based segmentation via the Cellpose software4 originally developed by Pachitariu and Stringer. My pipelines identify whole cells and nuclei from cells stained with fluorescent markers including DAPI (a nuclear label), Wheat Germ Agglutinin (WGA; labels cell membranes) and Sarcomeric α-Actinin (which stains both the cortex of the cell along with cardiomyocyte Z-discs). In the 2D pipeline, cell and nucleus segmentation are used to calculate cell shape, nuclear shape, and protein intensity at the single-cell level. After training, my algorithm could successfully parse the cortical component of Sarcomeres α-Actinin staining to segment cells. A major focus of the 2D analysis is NFAT localization between the nucleus and cytoplasm, since this reflects important signaling activity in cardiomyocytes. The 2D workflow also includes analysis of sarcomere organization.

The framework was also extended to 3D image analysis so that z-stack microscopy images could be processed across multiple slices instead of only a single plane. In the 3D workflow, segmentation was used to identify individual cardiomyocytes to determine whether increased expression of the intermediate filament protein, desmin, in 3D reflected greater expression per cell, or a higher percentage of positive cells. These technical advances made it possible to study not only protein intensity, but also how abnormal protein accumulation is distributed throughout the cell volume. By combining segmentation, protein quantification, and structural analysis, the framework provides a more complete view of cellular changes associated with cardiomyopathy.

ix

In addition, this thesis describes Cellpose fine-tuning, 2D and 3D image-processing workflows, and development of a Django-based application to make the analysis easier to use for non-experts. Overall, this work provides a more reproducible, scalable, and user-friendly approach for unbiased quantitative analysis of cardiomyocyte fluorescence microscopy images, with future potential for cloud-based biomedical image analysis.

Committee Chair

Dr.Huebsch, Nathaniel

Committee Members

Dr. Abhinav Kumar Jha Dr. Abhinav Diwan

Degree

Master of Science (MS)

Author's Department

Biomedical Engineering

Author's School

McKelvey School of Engineering

Document Type

Thesis

Date of Award

Spring 5-6-2026

Language

English (en)

Author's ORCID

https://orcid.org/0000-0001-6966-6685

Available for download on Thursday, October 22, 2026

Share

COinS