Purchase this article with an account.
Shuang Yu, Maryam Mehdizadeh, Shaun Frost, Di Xiao, Yogesan Kanagasingam; Graph Theory Based Intelligent Retinal Vessel Analysis. Invest. Ophthalmol. Vis. Sci. 2017;58(8):656.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
An intelligent retinal vessel analysis program is developed in this research. The program is based on graph theory and requires minimal user interactions for retinal vessel network analysis.
Automatic image processing algorithms have been developed to extract the vessel graph structure with procedures of optic disc detection, vessel segmentation, branch/crossover identification, root node detection and graph tree extraction. Afterward, arterioles/venuoles (A/V) are classified using Kmeans clustering and fine-tuned with the graph structure among vessel segments. User interaction is enabled at every step to rectify the automatic result.In order to make the system more intelligent and minimize user effort, a set of self-consistency checking rules are designed to automatically detect errors in the extracted vessel graph structure. The checking rules are established based on 3 assumptions for the correct vessel tree: 1, it has and only has one root node; 2, it can only contain bifurcation points and contain no crossover points within itself; 3, it contains no closed loops. The incorrectly extracted trees are automatically detected and highlighted, so that users can allocate their attention wisely to rectify the incorrect ones. After proper user intervention, vessel parameters are automatically calculated, including but not limited to central retinal equivalent calibers, A/V width ratio, fractal dimension, lacunarity, tortuosity, central reflex and branching parameters etc.The system was preliminarily tested on 14 disc centered color fundus images captured with a Canon CR-1 non-mydriatic camera at field of view.
Fig. 1B shows the automatically extracted tree graph for the image shown on 1A. The incorrect vessels are highlighted in white. 1C and 1D show the user rectified tree graph and A/V classification respectively. According to the preliminary test, the system is able to improve the A/V classification accuracy from 55.42% to 100% within 5 minutes of user interaction in average.
We have developed a graph theory based retinal vessel analysis system, which is able to provide intelligent hints on necessary user interventions. Therefore, users are enabled to correct the vessel structure with high efficiency and minimal interaction.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
Fig.1 A, Color fundus image; B, Automatically extracted vessel graph and incorrect trees are colored in white; C, Rectified vessel graph; D, A/V classification.
This PDF is available to Subscribers Only