The Impact of Fundus Autofluorescence on the Management of Age-related Macular Degeneration

Abstract

Background: Fundus autofluorescence (FAF) has been described as a topographical map of fluorophores that accumulate within the retinal pigment epithelium as a result of disease.
Study aims: To evaluate whether FAF offers information relevant to age-related macular degeneration over that gathered via colour fundus photography (CFP) and optical coherence tomography (OCT).
Methods: Ninety-three patients were imaged via CFP, OCT and FAF and the results analysed using Orange Data Mining artificial intelligence and SPSS software.
Results: Pupillary dilation makes a significant improvement to FAF image quality.
Nuclear sclerotic cataract of > 1.5 on the World Health Organisation scale indicates that there is ≃85% probability that the FAF image will not be of high quality. At > 1.9 there is ≃50% probability of the image not being clinically useful as defined by a novel grading scale.
Age was negatively associated with FAF comfort.
There is ≥ 90% probability of an abnormal FAF result for an eye with any of the following:
> 50 small, > 40 intermediate, > 20 large drusen. Age > 92 years. > 30 packet years of smoking. Any pigmentary abnormalities. ≃80% for any reticular pseudodrusen (RPD).
FAF results can be predicted via CFP and OCT data using machine learning with informedness of up to 70.2% and area under the curve (AUC) of 0.903. For transfer learning to be useful within primary care, image pre-processing is likely to be required.
Geographic atrophy and pigment epithelial detachments appear to be linked to a patchy FAF pattern. RPD are linked to a reticular FAF pattern.
Principle component analysis indicates that drusen were responsible for the greatest percentage of variability in this study’s data (38.6%).
Conclusions: Clinical impact: FAF results can be predicted from CFP/OCT via machine learning with 70.2% informedness and AUC of 0.903. Drusen number/size were the most informative variables.

Date of Award	Mar 2023
Original language	English
Supervisor	Hannah Bartlett (Supervisor)