Purpose: The retinal vascular tree exhibits fractal characteristics. These findings relate to the mechanisms involved in the vascularization process and to the objective morphologic characterization of retinal vessels using fractal analysis. Although normal retinas show uniform patterns of blood vessels, in pathologic retinas with central vein or artery occlusions, the patterns are irregular. Because the generalized box fractal dimension fails to differentiate successfully between normal and abnormal retinal vessels in 60 degrees fluorescein angiograms, the authors have further investigated this problem using the local connected fractal dimension (alpha).
Methods: The authors studied 24 digitized 60 degrees fluorescein angiograms of patients with normal retinas and 5 angiograms of patients with central retinal vein or artery occlusion. The pointwise method estimated the local complexity of the angiogram within a finite window centered on those pixels that belong to the retinal vessels. Color-coded dimensional images of the angiograms were constructed by plotting the pixels forming the object with a color that corresponded to specific values of alpha +/- delta alpha.
Results: The color-coded representation allowed recognition of areas with increased or decreased local angiogram complexity. The alpha distributions showed differences between normal and pathologic retinas, which overcomes problems encountered when using the methods of calculating the generalized fractal dimensions. A multivariate linear discriminant function using parameters from the alpha distribution and a further fractal parameter--lacunarity--reclassified 23 of the 24 normal and 4 of the 5 pathologic angiograms in their original groups (total: 92.1% correct).
Conclusions: This methodology may be used for automatic detection and objective characterization of local retinal vessel abnormalities.