Purchase this article with an account.
T.R. Friberg, M. Palaiou, L. Jefferson, B. Burtt; Drusen Detection: Precision and Accuracy of a Semi-Automated Method . Invest. Ophthalmol. Vis. Sci. 2003;44(13):1799.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To determine the sensitivity and accuracy of a semi-automated drusen detection program for use on images of eyes with age-related macular degeneration. Methods: Two independent research assistants with no experience in ophthalmology were given 2 hours of training in the use of a drusen detection computer algorithm (A.R.T.T., Crete, Greece; Iridex, Mt. View, California) which allows the operator to interact with the program until all viewable drusen are counted. The device displays the total drusen area in units of 0.01 mm2. After training, 14 digital images of eyes with drusen were evaluated to determine the total drusen area in a given circle of regard; measurements of all the images were repeated 10 times in a masked fashion. A retinal specialist also performed the analysis. The precision of the measurements was determined by calculating the standard deviation of the mean counts for each eye for each operator. The accuracy was determined by using the average of the 10 area measurements for each eye as determined by the ophthalmologist as the standard. After one week of training, the research assistants repeated their measurements. Results: The standard deviation (precision) of the drusen measurement process for the operators varied from between +10 to +115% of the mean area value as determined for each image. The average precision (n=10) was +28% and +57% respectively, while it was +22% for the clinician. The difference between the quantitative area measurements and the standard was +60% and +75% for each operator. The greatest errors occurred in poor quality images and in images in which the drusen were sparse. When the exercise was repeated after more extensive training, the average precision improved to +22% and +38% of the measured area value. The average accuracy (difference from standard) increased with the error differences decreasing to +45% and +46% of the standard value. All differences were statistically significant (P < 0.05). The time to analyze each image averaged 5-7 minutes. Conclusion: Quantitative measurements of drusen area can be performed quickly in a reasonably precise manner. The detection of drusen on any photograph or even on clinical examination is somewhat subjective. For this reason, training of the operator in drusen detection is important in order to achieve results which are consistent with a retinal specialist’s interpretation and to reduce the "noise" of the measurement. The speed at which drusen can be measured semi-automatically allows multiple measurements of each image to be made more practically, and this is desirable particularly when image quality is poor.
This PDF is available to Subscribers Only