Abstract
Purpose: :
To use fractal properties of AMD retinal images to classify AMD pathology.
Methods: :
The data set includes 30 images of normal eyes and 286 images of diseased eyes in the Macular Genetics Study at Columbia University with varying levels of age-related macular degeneration (AMD). Each image corresponds to a different patient. The diseased eyes were separated into 3 different stages of AMD (Early, Intermediate, and Advanced) according to the International Grading System. Each image was cropped with a 1000 pixel square centered on the macula comprising the central 6000 micron diameter circle, not including the optic nerve. The fractal dimensions of the images were obtained using the NIH ImageJ software, and their distributions were analyzed by t-tests.
Results: :
The fractal for a normal eye image seems to be higher than the fractal of an AMD image, regardless of the level of pathology. As AMD progresses, the fractal obtained from the image decreases and the standard deviation increases, consistent with the variable morphology in later stages of AMD (Table 1). The mean fractal of the normal images was significantly different from that of each AMD severity level. The fractals of the varying AMD severities were also significantly different from each other. (Table 2). A fractal dimension of 1.88 or lower identified macular disease of at least intermediate severity with 64% specificity and 70% sensitivity.
Conclusions: :
As AMD progresses, the fractal dimension of the image significantly decreases. Because higher fractal dimension generally corresponds to higher image complexity or information content, this suggests there is more information in the image of a better functioning macula, and that this information is more finely encoded than the grossly visible lesions of AMD. That this single fractal obtained without reference to usual structures differentiates AMD stages is remarkable. Further study is warranted to understand this relationship and how it might be applied to clinical assessment of patients.
Keywords: image processing • imaging/image analysis: clinical • macula/fovea