Abstract
Purpose :
We have developed and tested a fast web-based retinal vascular analysis program called VASP. This web-based database driven tool allows multiple users to access the service simultaneously to process retinal images and calculate vascular parameters. The parameters get stored in a central secure database for data analysis and identification of potential retinal biomarkers using machine learning techniques.
Methods :
Retinal images from various fundus cameras are imported, pre-processed and undergo automated detection of optic disc and vessel network. Vessels are automatically labelled as arteries and veins and bifurcations and cross-overs are classified. User input is obtained at multiple stages to correct errors in the automated algorithms. The generated parameters, intermediate results, and data relations including graph structures are stored in a centralized database for future data analysis of potential identification of retinal biomarkers or automatic testing of newly implemented techniques. Label propagation through the vessel network and graph structure is encoded in the data visualization techniques so that minimal user interaction is required to correct and adjust the vessel networks. The Dynamic data visualizations allow the user to adjust the vessel widths, and the largest six arteries and veins. A set of 22 color fundus images are used to test the system. Images are 1500x1152 pixels covering 45 field of view.
Results :
The average time to process images including user interaction is 4 minutes. Most of the diagnosis is done automatically. As displayed on the Fig.1, the system allows users to correct vessel network graph (Fig. 1(A)), adjust vessel width measurement (Fig. 1(B)) and edit largest artery vein vessels (Fig. 1(C)).
Conclusions :
The preliminary results demonstrate that the system is effective and intelligent to generate vessel parameters. Processing time, ease of use, and clinical effectiveness were our main focus to develop VASP. We are using VASP to analyze images from patients with hypertension and stroke.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.