Date on Master's Thesis/Doctoral Dissertation

8-2023

Document Type

Doctoral Dissertation

Degree Name

Ph. D.

Department

Computer Engineering and Computer Science

Degree Program

Computer Science and Engineering, PhD

Committee Chair

Elmaghraby, Adel

Committee Co-Chair (if applicable)

El-Baz, Ayman

Committee Member

Imam, Ibrahim

Committee Member

Park, Juw

Author's Keywords

retina; segmentation; diabetes; diagnosis

Abstract

This dissertation makes significant contributions to the field of ophthalmology, addressing the segmentation of retinal layers and the diagnosis of diabetic retinopathy (DR). The first contribution is a novel 3D segmentation approach that leverages the patientspecific anatomy of retinal layers. This approach demonstrates superior accuracy in segmenting all retinal layers from a 3D retinal image compared to current state-of-the-art methods. It also offers enhanced speed, enabling potential clinical applications. The proposed segmentation approach holds great potential for supporting surgical planning and guidance in retinal procedures such as retinal detachment repair or macular hole closure. Surgeons can benefit from the accurate delineation of retinal layers, enabling better understanding of the anatomical structure and more effective surgical interventions. Moreover, real-time guidance systems can be developed to assist surgeons during procedures, improving overall patient outcomes. The second contribution of this dissertation is the introduction of a novel computeraided diagnosis (CAD) system for precise identification of diabetic retinopathy. The CAD system utilizes 3D-OCT imaging and employs an innovative approach that extracts two distinct features: first-order reflectivity and 3D thickness. These features are then fused and used to train and test a neural network classifier. The proposed CAD system exhibits promising results, surpassing other machine learning and deep learning algorithms commonly employed in DR detection. This demonstrates the effectiveness of the comprehensive analysis approach employed by the CAD system, which considers both low-level and high-level data from the 3D retinal layers. The CAD system presents a groundbreaking contribution to the field, as it goes beyond conventional methods, optimizing backpropagated neural networks to integrate multiple levels of information effectively. By achieving superior performance, the proposed CAD system showcases its potential in accurately diagnosing DR and aiding in the prevention of vision loss. In conclusion, this dissertation presents novel approaches for the segmentation of retinal layers and the diagnosis of diabetic retinopathy. The proposed methods exhibit significant improvements in accuracy, speed, and performance compared to existing techniques, opening new avenues for clinical applications and advancements in the field of ophthalmology. By addressing future research directions, such as testing on larger datasets, exploring alternative algorithms, and incorporating user feedback, the proposed methods can be further refined and developed into robust, accurate, and clinically valuable tools for diagnosing and monitoring retinal diseases.

Share

COinS