Automated Analysis of Optical Coherence Tomography Angiography Images for Age-Related Macular Degeneration

Abstract

Age-related Macular Degeneration (AMD) is a damaging and threatening retinal condition and the leading cause of visual impairment in the elderly population. Optical Coherence Tomography Angiography (OCTA) is a relatively new imaging technique that enables the visualisation and characterisation of the retina of the human eye. The OCTA imaging technique produces clear images of the retinal and choroidal vascular layers including the superficial inner retina, the deep inner retina, the outer retina and the choriocapillaris layers. The current clinical standard for detecting and evaluating the efficacy of the treatments for AMD disease involves visually examining the textural appearance of OCTA images of these layers. However, this is not a trivial task given the significant amount of data being acquired in each OCTA scan, the pattern variations between individuals, and the fact that areas of abnormalities may appear similar. As a consequence, it may exceed the clinician's ability to visually diagnose AMD patients accurately. Therefore, this research aims at automating reliable ways of accurately quantifying and finding evidence of AMD presence in the context of OCTA images, as such evidence is not easily perceptible by ophthalmologists. Enabling automated analysis of OCTA images texture could also have a significant impact on ophthalmologists' workload. In this research, three different algorithms have been developed to help in quantifying, localising and classifying AMD disease in OCTA images in an automated manner. The first algorithm is constructed for conducting image classification based on whole local texture features as developed by testing different texture descriptors including the Local Binary Patterns (LBP) and the Binary Robust Independent Elementary Features (BRIEF) for measuring the texture of OCTA images and examining various classifiers including the Support Vector Machine (SVM) and the K-Nearest Neighbour (KNN) for performing the classification. The second algorithm is analogous to the first algorithm; however, it is based on reduced-local texture features as transformed by the Principal Component Analysis (PCA) technique to decorrelate the texture features. The third algorithm is created for localising areas of ocular vascular abnormalities related to AMD disease in the texture of OCTA images. This is accomplished by employing the LBP texture descriptor and testing different distance similarity metrics including the chi-square and the histogram intersection for performing the localisation. The various diagnostic algorithms developed were rigorously evaluated based on diverse OCTA image data sets provided by two different hospitals, namely the Manchester Royal Eye Hospital and the Moorfields Eye Hospital. Several eye conditions involved in the various OCTA image data sets comprised healthy, dry AMD and wet AMD. Broadly, the first algorithm demonstrated to perform best on classifying the healthy cases from the wet AMD disease cases accomplishing a mean area under the receiver operating characteristic curve (AUC) score and a standard deviation = 1.00 ± 0.00. The second algorithm, on the other hand, showed better performance on distinguishing eyes with dry AMD disease from eyes with wet AMD disease achieving a mean AUC score and a standard deviation = 0.85 ± 0.02, and on distinguishing eyes with choroidal neovascularisation (CNV) lesions related to AMD disease from eyes with non-CNV (dry AMD) lesions attaining a mean AUC score and a standard deviation = 0.85 ± 0.05. The third algorithm, however, was shown to provide reasonable estimates for the approximate locations of CNV regions in the diseased OCTA images of the outer retina and the choriocapillaris layers. Overall, the results obtained from the various algorithms developed are very effective and encouraging compared to other well-established methods. These algorithms have the potential to be integrated into the OCTA imaging technique, hence providing an insta

Date of Award	31 Dec 2022
Original language	English
Awarding Institution	The University of Manchester
Supervisor	David Morris (Supervisor) & Timothy Cootes (Supervisor)