Advisory Committee Members
Massimo A Fazio
J Crawford Downs
Date of Award
Degree Name by School
Doctor of Philosophy (PhD) School of Engineering
Approximately 80 million people in the world were estimated to suffer from glaucoma in 2020. To date, lowering intraocular pressure (IOP) is the only clinical treatment to delay or prevent the development and progression of glaucomatous vision loss. However, many glaucoma patients have IOP in the normal range, while some people with abnormally high pressure do not develop neuropathy. This indicates that biomechanical responses of the optic nerve head (ONH) to changes in IOP can vary significantly among individuals. This study investigated acute IOP-induced compliance in the ONH via SDOCT imaging, using computer vision, and deep learning networks to improve understanding of the association the morphological changes with glaucomatous neuropathy. First, developing models demonstrating the relationship between IOP and morphological changes in the ONH of living eyes were performed. To separate the systematic variability induced by IOP from the stochastic variability due to pulsatile and eye motions, we proposed a normalized histogram-based image registration approach utilizing image spatial information. As a result, discontinuities of A-scans and the tilt of the central axis were reduced. To determine if IOP-induced strains are a metric of acute morphological changes in various anatomical ONH tissues, we assessed the repeatability of in vivo strain measurements. Strain variability across eyes was greater than variation iv across imaging sessions, indicating that intersession strains of each eye are repeatable. On the current testing platform, strain and IOP were measured separately by two different devices. We synchronized the two measurements based on their strong correlation with the cardiac cycle under controlled test conditions. Preliminary data depicting the relation between IOP and strain indicate that models could be built by the optimization process. Second, convolutional neural network (CNN) based OCT scan segmentation with improved robustness was developed by simulating multiple degradation types of clinical OCT imaging to measure morphological changes in the ONH tissues. Preliminary data indicates that when segmenting OCT scans acquired in clinical settings, the performance of the newly trained models is improved. It implies that the CNN-based OCT segmentation approach with improved robustness can be deployed in clinical settings.
Kim, Jihee, "Dynamic Biomechanical Characteristics of the Optic Nerve Head by OCT Imaging" (2021). All ETDs from UAB. 633.