Purchase this article with an account.
Vinayak S Joshi, E Simon Barriga, Sheila C Nemeth, Wendall Bauman, Peter Soliz; Automated Assessment of Retinal Vascular Abnormalities for Computer-assisted Screening of Retinopathies. Invest. Ophthalmol. Vis. Sci. 2014;55(13):233. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To present a fully automated system that uses digital fundus images to detect retinal vascular abnormalities indicative of sight- or life-threatening conditions.
The automated retinal vascular abnormality detection system is developed using innovative algorithms for identification of abnormal vessel branching angles, abnormal branching coefficient, artery-venous (AV) nicking, copper/silver wiring, and retinal emboli. These algorithms are based on the vascular features describing color, intensity variation, and morphology; as well as the information obtained from vessel analysis techniques developed for vessel segmentation, segmentation correction, vessel separation, and artery-venous classification. After evaluating the images for quality, the system implements the automatic vessel analysis techniques, and detects the vascular abnormalities. The abnormality information can then be integrated with quantitative vessel morphology properties such as AV ratio and tortuosity, into a graphical user interface to aid the retinal grader in computer-assisted screening of retinopathies. A block diagram in Figure 1 shows the innovative methods in yellow and the previously developed methods in blue.
We tested each vessel abnormality detection algorithm with a dataset of 25 fundus images, resulting in an average accuracy of 77% against the retinal specialist’s ground truth. This is a marked improvement over previously reported accuracy of retinal vascular abnormalities detection. This software can enhance retinal graders performance by: 1) Improving grader agreement in detection of retinal vascular abnormalities (currently reported Kappa=0.4); 2) Reducing the required grading time by obviating the need for manual analysis (current grading time/patient=15 min).
This work presents the development of an automated system for detecting retinal vascular abnormalities, which enhances grader’s screening ability in terms of accuracy, grading time, and consistency. This system will aid in improving the screening of retinal vascular abnormalities that are indicative of sight- or life-threatening conditions such as stroke or cardiovascular infarction.
This PDF is available to Subscribers Only