Purchase this article with an account.
David Meadows, Nicholas Dunbar, Floyd Turner, Jacob Panikulam, Jessica Lim, Daniel Lehnen; Detection of vascular diseases in the retina using Constructal biomarkers to access fundus photographs. Invest. Ophthalmol. Vis. Sci. 2016;57(12):1710.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To determine if retinal vascular biomarkers based on a Constructal biomimicry framework are robust metrics to screen for ocular and systemic diseases when computed from the morphology and blood flow characteristics of retinal arteries and veins.
A new software framework has been developed for automatically classifying blood flow capacity of retinal vascular networks into either healthy (H), glaucoma (G), diabetic retinopathy (DR) or hypertension maculopathy (HM) categories using fundus images. This system (OQULUSTM) uses mathematical Constructal biomimicry principals to provide quantitative biomarkers of the retinal vasculature. High-resolution 50o fundus images were used for H, G, DR and HM patients from open source and proprietary image databases. Images were segmented to isolate and classify retinal arteries or veins. Biomarkers were identified for both mean and median arterial and venous networks using univariate and multivariate ROC curves. Thresholds based on Youden Criteria and Principal Components Analyses (PCA) were performed with and without Varimax rotation. Voronoi analyses were performed on each vascular network yielding maps of blood flow density indicating regions of poor blood perfusion.
Large differences existed in overall network blood flow rates between H, G and DR patients, see Table. Univariate ROC and multivariate PCA identified a wide array of biomarkers specific for each disease that achieved excellent precision. These factors primarily correlated with features affecting blood flow and vascular structure. Voronoi flow density analyses were capable of identifying DR patients with varying degrees of neovascularization, see Figure.
The Constructal framework provides a robust mathematical basis for selecting disease-specific biomarkers that enable automated screening for multiple diseases in patients from a single fundus image. These metrics highlight the impact that various diseases have on structural and blood flow characteristics of the retina vasculature. In G patients, there is a global narrowing of the arterial and venous networks. In DR, patients there is significant arterial small-vessel dropout that diminishes the functionality of the vascular network. In HM patients, increased vessel tortuosity and vessel narrowing caused decreased blood flow.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
Examples of ROC Analyses
Voronoi analysis for DR patient
This PDF is available to Subscribers Only