london euretina
retina race

Automated classification of retinal blood vessels in confocal laser scanning microscopy fundus images

Session Details

Session Title: Imaging III

Session Date/Time: Saturday 19/09/2015 | 16:30-18:00

Paper Time: 17:10

Venue: Athena.

First Author: : R.Kromer GERMANY

Co Author(s): :    A. Bartels   M. Klemm           

Abstract Details

PURPOSE:First pathologic alterations of the retina are seen in the vessel network. These modifications affect arteries and veins very differently. In order to develop an automatic procedure for the diagnosis and grading of retinopathy, it is necessary to be able to discriminate arteries from veins. There have been a few researches for classification of retinal blood vessels for conventional fundus photography images and this work focuses on the processing of confocal scanning laser ophthalmoscopy (cSLO) images, which are grayscale images.

Setting:

33 cSLO images from 30 healthy patients taken with „Spectralis' (Heidelberg Engineering, Heidelberg, Germany), followed by manual classification for ground truth by one ophthalmologist.

Methods:

The algorithms were written in the commercially available software package „Matlab“ (MathWorks Inc., USA). The automated classification is done by vessel segmentation, optic disc detection, estimation of the disc diameter, thinning of vessel segments for a skeleton, spline approximation for perpendicular intensity profile calculation to the vessel direction and a piecewise Gaussian model for the feature extraction. The features extracted by using the Gaussian model are then used to train a support vector machine classifier and by evaluating different kernel functions and features. These results were improved by developing and implementing an algorithm for junction detection and thus forcing all vessel segments along one vessel structure to be limited to one class, either artery or vein.

Results:

A total of 603 segments of arteries and 440 segments of veins were tested by the proposed algorithm. Classification had a sensitivity of 87,12% and a specificity of 74,77% for differentiation between different vessel classes.

Conclusions:

This shows that it might be possible to automatically analyze and classify arteries and veins in cSLO retina images. The results might be improved by facilitating vessel tracking in order to detect vessels that cross and run parallel.

Back to previous
EURETINA, Temple House, Temple Road, Blackrock, Co Dublin. | Phone: 00353 1 2100092 | Fax: 00353 1 2091112 | Email: euretina@euretina.org

Privacy policyHotel Terms and Conditions Cancellation policy