Deep Learning Image Analysis Framework for Clinical Management of Retinal and Corneal Diseases

Limited Access
This item is unavailable until:
2024-09-16

Date

2022

Journal Title

Journal ISSN

Volume Title

Repository Usage Stats

110
views
0
downloads

Abstract

Retinal and corneal diseases are the leading causes of global vision impairment. The advent of high‐resolution ophthalmic imaging technology enables the visualization of internal and external structures of the eye, thereby aiding clinicians in the assessment of these diseases. However, there are high costs and constraints associated with manual image analysis by clinicians. Therefore, the development of automatic algorithms with suitable performance for clinical practice is crucial to alleviate the burden on clinicians and improve the efficiency of the clinical workflow. This dissertation describes the development of a deep learning‐based image analysis framework consisting of computational algorithms for accurate automatic medical image analysis applied to ophthalmology for the clinical assessment of retinal and corneal diseases.

In Chapter 2, we developed longitudinal algorithms for the assessment of 3‐D medical images with applications for optical coherence tomography (OCT). For clinical application, we used the algorithms for the assessment of macular telangiectasia type 2 and USH2A‐related retinitis pigmentosa in large‐scale clinical trials. We also introduced the concept of longitudinal transfer learning to develop personalized algorithms and introduced a new paradigm for validating automatic algorithms for clinical applications beyond performance metrics.

In Chapter 3, we developed region‐based algorithms for the assessment of 2‐D medical images with applications for slit lamp photography (SLP). For clinical application, we used the algorithms for the assessment of microbial keratitis in clinical studies from the USA and India. We also demonstrated the potential of automatic SLP-based measurements for assessing ocular function and paved the way for the development of objective and standardized strategies for the assessment of corneal diseases.

In Chapter 4, we developed hybrid algorithms for the joint assessment of spatially-registered 2‐D and 3‐D medical images with applications for indocyanine green angiography and OCT. For clinical application, we used the algorithms for the assessment of polypoidal choroidal vasculopathy, a sub‐type of age‐related macular degeneration. We introduced hybrid network architectures with fusion attention modules that effectively processed co‐registered images of different dimensionalities to enable sharing of learned features between the different imaging modalities. We also derived quantitative definitions of important imaging biomarkers of the disease.

In conclusion, this dissertation provides an image analysis framework for clinical management of retinal and corneal diseases. Our deep learning‐based computational algorithms can accurately identify and quantify important disease biomarkers automatically and have been validated for several aspects of clinical applicability. These algorithms can be used by clinicians to improve the efficiency of the clinical workflow, leading to timely and precise medical decisions, ultimately improving patient outcomes.

Description

Provenance

Citation

Citation

Loo, Jessica (2022). Deep Learning Image Analysis Framework for Clinical Management of Retinal and Corneal Diseases. Dissertation, Duke University. Retrieved from https://hdl.handle.net/10161/25750.

Collections


Dukes student scholarship is made available to the public using a Creative Commons Attribution / Non-commercial / No derivative (CC-BY-NC-ND) license.