Loading...
Using machine learning to predict f-actin morphology of endothelial cells: An application for mechanobiology models
Hafenstine, Rex W.
Hafenstine, Rex W.
Citations
Altmetric:
Files
Loading...
thesis
Adobe PDF, 1.33 MB
Authors
Other Names
Location
Time Period
Advisors
Original Date
Digitization Date
Issue Date
2023-05
Type
Thesis
Genre
Keywords
Subjects (LCSH)
Electronic dissertations
Citation
Abstract
The expansive monolayer of cells in direct contact with blood is called the endothelium. The endothelium is fundamentally involved with almost every human disease. The endothelium is composed of endothelial cells (EC)s that exhibit structural and phenotypical heterogeneity that vary in time and space. These cells exhibit emergent properties that allow communication with neighboring cells. Since experimental observation of ECs at a single cell level will not provide complete details on these behaviors, other techniques are needed. One method for observing these emergent properties is through 3D dynamic response of cell shape and morphology. Live-cell imaging is limited, such as the number of structures or events that can be imaged. Therefore, incorporating a complementary method such as machine learning (ML) could be a feasible option. To validate if ML could predict 3D f-actin fibers from only the cell membrane, human dermal microvascular endothelial cells were grown to confluence, the cell membrane, nucleus, and f-actin were fluorescently labeled, and imaged with confocal microscopy. The images were processed via normalization techniques, segmented, augmented, and filtered before being introduced into a conditional generative adversarial network (cGAN). The f-actin predictions from the cGAN did not perform at a level previously seen when predicting 3D nucleus and focal adhesion (FA) structures. However, these results do not necessarily mean that the f-actin fibers cannot be predicted but may require different methods, such as tuning the cGAN parameters (batch size and additional learning rates), obtaining more raw images, or testing different deep learning (DL) algorithms. Future work should also include testing a primary cell line of human dermal blood endothelial cells to minimize cell overgrowth, transfecting the cell with a K-Ras CAAX motif to improve cell membrane labeling, and developing new image similarity metrics to compare the predicted f-actin images with their corresponding ground truth images.
Table of Contents
Description
Thesis (M.S.)-- Wichita State University, College of Engineering, Dept. of Biomedical Engineering
Publisher
Wichita State University
Journal
Book Title
Series
Digital Collection
Finding Aid URL
Use and Reproduction
© Copyright 2023 by Rex W. Hafenstine
All Rights Reserved