Purpose:
To develop a computerized system to quantitatively assess the vascular parameters in retinal images from normal, hypertensive or diabetic subjects. It will allow clinicians to recover the quantitative estimation of these parameters in an objective, fast and user-friendly way.
Methods:
The first, totally automated phase performs image luminosity equalization and contrast enhancement, tracing of the vascular network with vessel caliber estimation, vessel classification as either arteries or veins. In the second phase, the user evaluates the results provided and confirm them or apply the needed corrections via a user-friendly editing interface, which is capable of detecting the troubled situations to be presented to the user for expert assistance. The vascular parameters are finally computed. Repeatability and accuracy of the complete system were assessed by having three trained users independently analyze 20 DCCT and 18 ETDRS images, kindly provided by the Dept. of Ophthalmology and Visual Sciences, University of Wisconsin, USA, whose IVAN program results were used as reference for accuracy.
Results:
The overall repeatability between users and accuracy with respect to IVAN results were assessed by the correlation coefficients, whose ranges over the 3 users are reported in the accompanying table. The total time required to analyze one image was on average 5 min (range 2-8 minutes).
Conclusions:
A computerized system for the fast and reliable quantification of vascular parameters CRAE, CRVE and AVR in retinal images has been developed and evaluated and will be soon available for general use.
Keywords: imaging/image analysis: clinical • retina • diabetic retinopathy