Optic disc pallor and brain health

Abstract

BACKGROUND: There is a growing need for accurate and cost-effective biomarkers for neurological disorders. Brain imaging is not suited to mass screening, owing to its cost and complexity of operation. The retina offers a promising and accessible alternative as it is directly connected to the brain via the optic nerve and shares its blood supply, embryological origins, and structural features.

Changes in the brain are often reflected in the retina. One such change observed in multiple neurological disorders is thinning of the retinal nerve fibre layer (RNFL), which is measured using Optical Coherence Tomography (OCT), reflecting loss of nerve tissue in the brain. However, OCT is not universally accessible and may be challenging to use with certain populations, such as frail individuals or those with movement disorders. An alternative approach still utilising the retina involves assessing the paleness of the optic disc in colour fundus photographs, based on the premise that a pale disc indicates RNFL degeneration or loss. Fundus photographs are the most widely available retinal imaging modality and routinely captured during high-street eye examinations and other standard clinical settings.

OBJECTIVE: In this thesis I aim to develop a method for quantifying optic disc pallor in fundus images, and to investigate optic disc pallor and in conditions affecting the health of the brain.

METHODS: I developed software to measure optic disc pallor using a combination of image processing techniques and deep learning segmentation networks. Validation was performed by comparing pallor measurements to peripapillary RNFL thickness from the same eyes and by assessing how pallor values aligned with clinically diagnosed pallor in an external dataset. I applied the software to three clinical datasets using linear, logistic, and linear mixed-effects models to investigate associations between pallor measurements and clinical features, including MRI-detected lesions, cerebral small vessel disease (cSVD) ratings, stroke type, and Parkinson’s disease (PD) status and duration. I also manually reviewed 1,485 images from an aggregated dataset and used these labels to create two methods for automated quality control.

RESULTS: The resulting software can process an image in under one second. In clinical research, I found that disc pallor was associated with MRI-visible enlarged perivascular spaces (indicative of underlying cSVD), lacunar stroke, cSVD ratings, and PD status and disease duration.

CONCLUSION: Optic disc pallor, measured in colour fundus photographs, reveals insights into neurological dysfunction. Future investigation is warranted to see if pallor could be an imaged-derived biomarker of disease or whether it holds predictive value in addition to conventional risk factors used clinically.