Abstract
Purpose: :
To compare a Graphical User Interface (GUI) with human graders and determine whether a GUI would be helpful in analyzing drusen in patients with AMD.
Methods: :
Digitized color fundus images of 167 AMD eyes (85 OD, 82 OS) with soft drusen from the Columbia Macular Genetics Study were analyzed by both a human grader and a GUI. The user-GUI interaction consisted of clicking the macular centre and a peripapillary point, then drawing the area the GUI had to search for drusen. The GUI centered, cropped, color balanced and resized the image, then calculated the area (percentage) of drusen for the centre circle (macular centre through 1000 microns), the inner ring (1000-3000 microns) and the outer ring (3000-6000 microns). Human graders estimated the percentages from the raw images in categories (0%, 0-10%, 10-25%, 25-50%, 50-100%). Results were analyzed by weighted kappas and sources of disagreement investigated
Results: :
Of the 167 images, agreements between the GUI and the human graders were 86 for the centre circle, and 102, and 120 for the inner resp. outer rings. The weighted kappa for the centre circle was 0.47 for OD and 0.50 for OS. For the inner circle OD and OS weighted kappas were 0.56 and 0.42, and the outer circle kappas were 0.42 and 0.34. Most disagreements were false positives from the GUI: geographic atrophy (GA), hypopigmentation, prominent choroidal vessels, or drusenoid pigment epithelial detachment (PED). Other sources were centration of the centre circle or small percentage disagreements causing categorical shifts. After the GUI further eliminated lesions causing false positives such as GA, the weighted kappa’s for the centre were OD=0.70, OS=0.66, inner circle: OD=0.73, OS=0,56 and outer circle OD=0.48, OS=0,43.
Conclusions: :
A GUI can be very useful for drusen analysis in patients with AMD. When implemented in user-friendly software the GUI can reduce time and increase accuracy for analysis of drusen in the different macular regions. For maximum accuracy, experienced human supervision is still necessary to remove false positive lesions .
Keywords: age-related macular degeneration • imaging/image analysis: clinical • retina