Student Theses, Dissertations and Capstones
Document Type
Thesis - NSU Access Only
Degree Name
Master of Science (M.S.) in Clinical Vision Research
Copyright Statement
All rights reserved. This publication is intended for use solely by faculty, students, and staff of Nova Southeastern University. No part of this publication may be reproduced, distributed, or transmitted in any form or by any means, now known or later developed, including but not limited to photocopying, recording, or other electronic or mechanical methods, without the prior written permission of the author or the publisher.
First Advisor
Bin Zhang
Publication Date / Copyright Date
1-1-2025
NSUWorks Citation
Austin Murr. 2025. Computational Methods to Optimize Orthokeratology for Myopia Control. Master's thesis. Nova Southeastern University. Retrieved from NSUWorks, . (18)
https://nsuworks.nova.edu/hpd_opt_stuetd/18.
Abstract
Background: Myopia rates are rising quickly worldwide and are expected to reach 50% of the global population by 2050. Developing treatments that can effectively slow the progression of myopia are therefore essential in the coming decades. One of the most promising myopia control treatments is orthokeratology (ortho-k), which uses rigid contact lenses to reshape the cornea. Parameters of the ortho-k treatment zone (TZ) are highly correlated with the treatment outcome. Though it can regularly halve axial length growth (ALG) in patients, ortho-k’s mechanism of action still remains unclear, making optimization difficult. Purpose: This thesis presents three computational methods that can be used in further research to better understand how ortho-k works and how it can be optimized. Chapter 3 introduces a new clustering algorithm (CBABC) that can reliably identify the ortho-k TZ in a standardized way. Chapter 4 expands upon the Riverbank model of the relative corneal refractive power shift (RCRPS), and presents an implementation of differential evolution to fit data to that model. Chapter 5 describes a generalized version of capsule networks (GenCaReNet) that can both predict ALG and provide interpretable information about which higher order aberrations (HOAs) are most influential in determining ALG outcomes. Methods: The three computational approaches were applied retrospectively to data from 200 subjects that fit with Euclid Emerald ortho-k lenses at Tianjin Medical University Eye Hospital (Tianjin, China). Feature data was extracted from Medmont E300 corneal topography data and raw ALG data was collected using the Lenstar 900. Results: The y-component of decentration was the TZ parameter most correlated with ALG (0.20, p=0.004**), and had a median (IQR) of -0.20 (-0.43 – 0.03) mm. For the Riverbank model, the location shift (0.23, p < 0.001***) and the new definition of slope (- 0.20, p = 0.005**) showed significant correlations with ALG. The GenCaReNet was able to predict ALG with impressive accuracy, attaining and RMSE of just 0.0970mm. Conclusion: The CBABC algorithm demonstrated it was capable of identifying the TZ autonomously in the vast majority of scenarios, and reinforced previous findings that TZ decentration is significantly correlated with ALG outcomes. Using an expanded version of the Riverbank model, it was shown that different model parameters are more correlated with ALG in different corneal octants. Additionally, the proposed algorithm suitably fit the RCRPS to the Riverbank model in every scenario. Lastly, the GenCaReNet was remarkably accurate in predicting ALG using just changes to the pupillary wavefront aberration as input. GenCaReNet’s reconstructive capabilities provide a fascinating—though yet unproven—capability to identify the aberrations most influential in determining the ALG, which revealed connections to the TZ and the Riverbank model of RCRPS. It is hypothesized that quinary comatic aberrations are important contributors to slowing the rate of ALG.
Disciplines
Optometry
Keywords
axial length growth; capsule network; computation; myopia; orthokeratology; treatment zone
Files
Download Full Text (2.7 MB)