Abstract
Purpose: :
Current methodology for the extraction of arteriovenous ratios (AVRs), a metric for hypertensive retinopathy, from fundus images requires time intensive, manual identification and measurement of vascular features. We have developed a set of image processing algorithms able to perform such analyses automatically on large databases of fundus images, offering significant diagnostic and screening potential. However the validity of our automated measurements is yet to be verified. Here we compare AVRs obtained by the automated software to manual calculations.
Methods: :
98 fundus images from the eyes of 49 patients were randomly selected from a database of hypertensive and non hypertensive patients. Vessels were classified as either artery or vein, vessel thickness measured, and AVRs calculated by the software and by a human observer. Paired t-test was performed to evaluate differences in manual and automated measurements.
Results: :
98 fundus images were analyzed. Between 34 and 76 vessels per eye were identified and measured, with strong agreement demonstrated between automated and manual results. There was no statistically significant difference in AVRs calculated by the software and human observers.
Conclusions: :
AVRs obtained from fundus photographs by automated software were shown to be strongly concordant with manual measurements. Automated AVR analysis of fundus photographs is a valid technology with promising applications.
Keywords: image processing • imaging/image analysis: clinical • retina