Abstract
Purpose: :
The association between retinal microvascular abnormalities and macrovascular diseases is partially represented by the artery–to–vein diameter ratio (AVR). To evaluate AVR–values gained by an automated system.
Methods: :
We developed an automated retinal vessel analysis system (ARVAS). It consists of 4 steps: (1) determination of the measurement zone, (2) vessel recognition, (3) vessel classification in arteries or veins, (4) calculation of AVR. The type of the retinal vessel, the position of measurement, and the diameter were obtained automatically. ARVAS allowed manual control and intervention of the suggested vessel classification: the type of a vessel could be changed or a vessel could be deleted. The AVR–values by ARVAS were compared with the AVR–values gained by the semiautomatical system used in the ARIC–study (ARIC system). By both methods the AVR–values were calculated with the scale dependent Parr–Hubbard formula and the scale independent Parr–Hubbard–Knudtson formula. 212 retinal digital photographs captured by the 45°–KOWA nonmydriatic fundus camera were evaluated by one investigator DB.
Results: :
By ARVAS, 49 of 212 images were automatically excluded from the calculation. In 46 images no veins could be identified and in 3 images the optic nerve head could not be detected. ARVAS required on average 4.8 changes from artery to vein and 0.3 changes from vein to artery per image by the investigator. The mean number of deleted vessels per image was 3.1 as these vessels were incorrectly segmented. To calculate AVR, ARVAS used in average 10.43 artery segments and 7.97 vein segments per image. The AVR–values gained by ARVAS and by ARIC differed significantly. The scale dependent AVR (Parr–Hubbard) by ARVAS and ARIC was in average 0.77±0.14 and 0.81±0.09, respectively. The scale independent AVR (Parr–Hubbard–Knudtson) was 0.61±0.1 and 0.68±0.07, respectively.
Conclusions: :
The application of ARVA in fundus photographs relieves the investigator of manual identification of the optic nerve head and vessels. Fully automated calculation of AVR is currently not possible because no reliable approach was found that would classify vessels into arteries or veins. Nevertheless, ARVAS improves reliability and reproducibility since the user can only change the label of vessels.
Keywords: imaging/image analysis: clinical • clinical (human) or epidemiologic studies: systems/equipment/techniques • retina