Fractal Characteristics of Retinal Microvascular Network in Alzheimer's Disease and Colon Cancer in Automatically Segmented Fundus Images from the UK Biobank Database

Abstract

Several studies have found that in people with cancer, the incidence of Alzheimer’s dementia (AD) is very low, while in people with AD the incidence of cancer is very low. These findings imply that both diseases develop through deregulation of a common pathophysiological process, which in each disease occurs in the opposite direction. In this study, we aimed to see if these differences are reflected by changes in retinal microvascular network complexity. Remodeling of retinal microvasculature has already been explored as a potential biomarker for AD. Although some hereditary forms of colon cancer (CC) are associated with retinal microvascular abnormalities, these changes have not been extensively studied. Three groups of retinal fundus images from the UK Biobank were selected, age and gender-matched: group with AD (n = 20), group with CC (n = 20), and a control group (CG, n = 20, participants without either of these two diagnoses). Blood vessels were segmented using a prediction model based on convolutional neural network. Thirty-eight (67.3%) images were successfully segmented. Box-counting dimension (Db) - a measure of microvascular complexity, and lacunarity (Λ) - a measure of microvascular gap distribution, were calculated. Db was lower in AD compared to CC and CG combined (mean Db, AD vs. CC+CG = 1.42 vs. 1.44, p = 0.048). Mean Λ was not different among the groups. More research is needed to improve automatic segmentation. Analysis of a larger sample that accounts for other known factors affecting microvascular complexity such as diabetes mellitus and hypertension should be done in future.