Abstract
Purpose:
Approximately 50 million fundus photographs are taken annually in the US and is expected to grow at an increasing rate. To address the growing burden on specialists and the increasing need for faster evaluation of these photographs, a new automated screening tool was developed which analyzes the blood flow performance of the vascular network observed in fundus images. This study investigates the ability of the methodology to identify new, quantitative imaging biomarkers for ophthalmic disease.
Methods:
We used 54 fundus photographs from the High-Resolution Fundus Image Database at the University of Erlangen-Nuremberg (http://www5.cs.fau.de/research/data/fundus-images/) that are fovea-centered with a field-of-view of 60 degrees. Each photograph is categorized as healthy or diabetically retinopathic. For every photograph, there was an accompanying manual segmentation of the vasculature from which we identified and separated the arterial and venous networks. The networks were analyzed to determine vessel geometries as well as vessel connectivity, which was used to calculate the total fluid conductance of the network through a series of linear algebraic operations. The apparent total blood flow was calculated for each network using a constant pressure gradient between the central vessel and capillaries. We compared the geometric, morphological and flow-related properties of arterial and venous networks from 13 healthy patients to those of 14 retinopathic patients.
Results:
The average apparent total blood flow in healthy arterial networks was 64.9 +/- 13.7 microliters per minute while the average for retinopathic arterial networks was 25.6 +/- 13.5 microliters per minute (p<0.001). There was no statistically significant difference between healthy venous flow and diabetic venous flow. There was a strong correlation between apparent arterial blood flow and total observable volume of arterial vessels (R=0.91) and a moderate correlation between apparent venous blood flow and total observable volume of venous vessels (R=0.57).
Conclusions:
Geometric and morphological characteristics of retinal vasculature provide quantitative imaging biomarkers for ophthalmic diseases. Apparent arterial blood flow is a significant indicator for the presence of diabetic retinopathy.
Keywords: 499 diabetic retinopathy •
549 image processing •
436 blood supply