Purchase this article with an account.
Ngan Meng Tan, Jiang Liu, Damon Wing Kee Wong, Jun Cheng, Carol Y. Cheung, Mani Baskaran, Tin Aung, Tien Y. Wong; An Evaluation for an Automated Left and Right Eye Identification System for Digital Fundus Images for Glaucoma Diagnosis. Invest. Ophthalmol. Vis. Sci. 2012;53(14):649.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To evaluate the performance of an automatic system to discern left and right eyes from non-stereoscopic digital fundus images.
We developed an automatic system to distinguish between left and right eyes in digital retinal fundus images. This distinction is characterized/motivated by the inclination of the central retinal vessels within the optic nerve head (ONH) towards its nasal region. A statistical approach using active shape model was used to segment the optic disc and a local adaptive filter was convolved with the optic disc to extract the central retinal vessels. From the binarized central vessel map, the ISNT regions of the ONH can be recognized. The ability to differentiate the ISNT regions will increase the domain knowledge for analyzing abnormalities in retinal images. The system was tested on 650 images (3072 x 2048, Field 2) randomly picked from the Singapore Malay Eye Study, and entered into ORIGA-light (Zhang et al, EMBC 2010), an online retinal fundus image database for glaucoma analysis and research. These retinal images were manually annotated by trained graders from the Singapore Eye Research Institute.
The 650 fundus images consist of 325 retinal images of each eye. The proposed automated system was able to correctly identify left and right eyes with a combined accuracy of 98% (637/ 650). Of the 13 images incorrectly identified, 5 images had peripapillary atrophy (PPA), 2 images had glaucoma, 2 images had media opacity and 4 images were normal.
The automatic system was able to identify left and right eyes from retinal images with 98% accuracy. Thus, the proposed system will allow application of the ISNT rule for glaucoma diagnosis. These resultshighlight the use of this system as an invaluable tool for other image segmentation and feature extraction methods in retinal fundus imaging.
This PDF is available to Subscribers Only