Purchase this article with an account.
M Wilson, SC Nemeth, P Soliz; An Evaluation Of A Computer-aided Diagnosis System For Segmenting Drusen And Calculating Their Distribution Statistics . Invest. Ophthalmol. Vis. Sci. 2002;43(13):2567.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose:To verify the accuracy of a computer-based segmentation algorithm for detection of drusen in subjects presenting with age-related macular degeneration by quantifying drusen size distribution statistics. The goal of this project is to improve the quantitative information produced when grading retinal images. Methods:Twenty-one retinal images from the «Wisconsin Age-related Maculopathy Grading Scheme» (WARMGS) were digitized. The images included hard and soft distinct and indistinct drusen as well as reticular drusen. The segmentation algorithm computed the locations of the drusen through a four-step process. The image is first filtered for lighting artifacts and noise. A threshold is then applied to multiple regions of interest to create areas where the probability of a drusen is high. Local maximum within these regions are taken as seed points and grown with a contrast-driven growing algorithm. The user is allowed to alter probability and seeding thresholds for addendum. Once the drusen are found the drusen size distributions are calculated. Applying the WARMGS protocol, location and size of the largest drusen in each of the nine WARMGS fields was automatically recorded. A comparison with the human-observer determined size was performed. A consortium of ophthalmologists at the University of Wisconsin has adjudicated the hand calculations. For all images, each of the 9 WARMGS grid locations were separately verified. Results:Size measurement accuracy for the computer ranged from correctly calculating 9 out of 9 for all nine regions to 5 out of 9. The size measurements were within 57 microns of accuracy for 89% of the images. Conclusion:The described computer-aided system shows promise in accurately determining the location and size of drusen in digital fundus images. This technology will aid in following drusen changes in drug and natural history studies.
This PDF is available to Subscribers Only