Purchase this article with an account.
B. Madjarov; Computer Assisted Grading Tool for Quantitation of Areas With Geographic Atrophy (GA) . Invest. Ophthalmol. Vis. Sci. 2006;47(13):2143.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To increase the grading accuracy from digital images through development and validation of robust algorithms for automated calculation of areas subtended by geographic atrophy in eyes with age related macular degeneration (AMD). Design and implementation of vendor–independent software platform to combine user–friendly graphical interface (GUI) and the calculation libraries in freely distributable package for the purposes of clinical trials.
Fast computer algorithms were developed using C++ programming language and tested on Pentium computer with 3.06 MHz Xeon processor. Outlining algorithm allows visualization of the borders of an area with GA on the digital image. Template algorithms facilitate digital placement of an electronic Wisconsin drusen template centered in the fovea. The software automatically calculates the area of GA . Multiple lesions may be selected and the areas are calculated separately and totaled in square millimeters. In order to decrease intra– and inter– operator variability between the measurements, an additional algorithm was developed to calculate automatically the shortest distance between the lesion border and the fovea. Calibration for different fundus cameras and viewing angles also was incorporated for consistency in the measurements. The results can be stored in a local customized database to permit later validation, comparison and data analysis. The outlined GA areas are saved separately from the original image for documentation and adjudication. An innovative GUI design allows for facile file retrieval, manipulation and screen interaction. The software is Windows based and runs on all versions.
The calculation algorithms were validated for accuracy. The area and linear calculations were tested against 100 different polygons and distances with known dimensions. The accuracy was 100% to the square pixel for areas and 100% to subpixel level for distances. Additional validation was performed on a set of 50 images with GA to assure that calibration, calculation, robustness and interface meet the study and reader requirements. Each image from a set of 10 images was regraded by the same operator and independently by two graders to assure reproducibility and measurement discrepancy of less than 5%.
A robust accurate vendor–independent application with facile interface was developed to allow for fast reliable computer–assisted grading of areas with GA on digital images.
This PDF is available to Subscribers Only