Abstract
Abstract: :
Purpose:Currently there is a lack of standardized topographic reference system of the fundus images for correlative and absolute measurements. We sought to develop and evaluate feasibility, reproducibility and accuracy of a global positioning system for fundus image grading to utilize in multicenter clinical trials. Methods: A custom designed software platform written in C++ programming language with graphical user interface was developed as a test environment. The application was evaluated on a computer with dual Pentium 4 processor and Windows operating system. One hundred digitized color fundus photographs at 1000 dpi were selected. An initial step was to identify the position of the fovea on the digital image. A coordinate system was automatically calculated and centered based on this localization. Each point from the digital image becomes a part of a new matrix represented by two values. A location in major quadrants is identified by unique combination of positive and negative values. A new measurement unit – fundus unit (FU) was introduced to eliminate the inaccuracies from the metric units. Evaluation was performed on the same image at different resolutions as well as on images from different fundus cameras with varying viewing degrees. Results: Evaluation on position accuracy based on multiple resolutions from the same image yielded 100 % identical coordinates in FU as judged by the same landmarks. Tests on images from the same fundus, photographed with two different fundus cameras revealed also 100 % identical positions. Locations of 10 arbitrary selected points confined to the main quadrants of the Wisconsin drusen grading template were tested on 100 images. There was a 99.6 % exact correlation between the positions of the points as judged according to their location to the grid. Tests results were 2% less accurate for panoramic image mosaic due to image manipulation and peripheral geometric and optical distortions. Conclusions:A new, accurate, robust global point identification system was developed to standardize the measurements and correlate the lesion location between different digital fundus image modalities.
Keywords: retina • imaging/image analysis: clinical • image processing