Microvascular network alterations in the retina of patients with Alzheimer's disease

Abstract

Patients with AD have altered microvascular network in the retina (narrower retinal venules and a sparser and more tortuous retinal vessels) compared with matched nondemented controls. These changes in retinal microvasculature may reflect similar pathophysiological processes in cerebral microvasculature in the brains of patients with AD.

Alzheimer’s disease: A review of its visual system neuropathology. Optical coherence tomography—a potential role as a study tool in vivo

“…In addition, one study using SD-OCT showed a diffuse reduction of the RNFL and GCL combined in AD [49], although the authors were not able to determinate which layer was most affected by AD. Later studies have demonstrated IPL thinning in AD patients [41,50,52,56,57]. This reduction of ganglion cell complex thickness (GC-IPL and RNFL layers) in AD occurs to a larger extent than that accounted for age-related GC-IPL loss alone (about 0.3 μm/ year) [54].…”

“…Laser Doppler studies have shown altered retinal blood flow in patients with AD, although no correlation was found between RNFL thickness and retinal blood flow [45]. Two recent studies using retinal photography found retinal vascular abnormalities, including vascular attenuation, reduced branching pattern, and altered venular tortuosity [54,56,57]. According to the authors, these retinal vascular changes in AD may be correlated with the neocortical amyloid plaque burden [58] or may mirror the cerebral microvascular pathology [54].…”

Retinal Biomarker Discovery for Dementia in an Elderly Diabetic Population

“…The parameter λ controls the strength of regularization. The parameter α is the elastic-net regularization mixing parameter; α = 1 corresponds to the lasso (also known as L 1 regularization) whereas α = 0 corresponds to ridge regression (L 2 regularization) [8]. For highly correlated features, the lasso tends to pick one of the features and discard the others, whereas the ridge shrinks the feature coefficients towards each other.…”

“…For highly correlated features, the lasso tends to pick one of the features and discard the others, whereas the ridge shrinks the feature coefficients towards each other. Elastic net mixes the two; for example, α = 0.5 tends to select or exclude groups of highly correlated features together [8].…”

“…The R package glmnet [8] was used to perform the regularized logistic regression experiments. glmnet implements an efficient algorithm for computing entire regularization paths, showing the effect of varying λ on the classification error and the number of features retained in the model.…”