Purchase this article with an account.
M. Palaiou, T.R. Friberg, R.W. Bremer, L. Jefferson, B. Mack; Automated Drusen Area Measurements Obtained From Digital Images versus Manual Reading Center Measurements Using Source Slide Material . Invest. Ophthalmol. Vis. Sci. 2005;46(13):1565.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To determine the correlation of macular drusen area measurements generated by manual readings conducted by an established reading center, to those generated using an automated drusen area measurement algorithm (Drusen AnalyzerTM, Iridex–A.R.T.T.) on digital images of eyes with age–related macular degeneration (AMD). Methods: A sample of 1205 hard copy Kodak Ektachrome slides were randomly selected from thousands which were obtained during the course of the Age–Related Eye Disease Study (AREDS) at one center. These were manually read by trained readers adhering to a rigorous protocol in which circular sizing templates and regional grids were utilized. The results were categorized by a reading center into 8 drusen area categories which ranged from 0.000 to 0.003 mm2 for category 0 eyes to more than 1.767 mm2 for category 7 eyes. Drusen area was measured in two concentric fields centered at the fovea with diameters of 1000 and 3000 microns respectively. Extra hard copy slides of the macula taken at the same sitting with the same camera were digitized using a Nikon Super Coolscan digitizer which rendered the image into 7–megabyte master TIFF files. The digitized images were then evaluated by a trained technician using the Drusen AnalyzerTM program in which autodetection of drusen was performed. The resultant readings from the algorithm yielded a continuous variable of drusen area in mm2. These readings were then converted to the categorical ranges used by the reading center so that a direct comparison could be made. The reading methods differ in several respects including the use of fixed–size templates regardless of apparent disc area in the manual method and the fact that every drusen is read individually and quantitated in the automated method. Results: The Spearman correlation coefficients for drusen area readings for the central 1000 and 3000 microns for the 1205 eyes were 0.63 and 0.66 respectively with p–values less than 0.0001 for each. This represents a categorical agreement of approximately 80% between the two methods. There is, however, no clear standard drusen detection method, as interpretation of what is and what is not a druse is ultimately somewhat subjective. Conclusions: Computer algorithm assisted drusen area measurements from digital images is a viable and likely more cost–effective method than using manual techniques. Such considerations may be pivotal, particularly if quantification of the risk of visual loss for an individual eye is based on morphology alone.
This PDF is available to Subscribers Only