Fractals for Complexity Analysis of Diabetic Retinopathy in Retinal Vasculature Images

Arterial pattern and morphology of distribution is damaged because of diabetes resulting in retinal vasculature deformation. This aspect is studied in terms of quantification of the degree of complexity associated with the distribution of blood vessels in eye for healthy and diabetic humans. Retina images of fifteen healthy subjects are compared with those of diabetic subjects. It is found that the increased complexity of structure and texture of the diabetic subjects results in a higher fractal dimension as compared to those of healthy subjects. Also the blood vessel patterns, both for healthy and diabetic subjects show self-similarity and scale invariance and hence the patterns are fractals. For the purpose of characterization of the irregular patterns of blood vessels in retina, box counting technique is used for the estimation of fractal dimension. A GUI based program is developed in Matlab for implementation of box counting technique and determination of fractal dimensions. Fractal dimension of retina images for diabetic subjects show higher fractal dimensions indicating higher degree of structural complexity associated with the image whereas images of healthy subjects show a lower value of fractal dimension indicating limited complexity of structure. It is shown that fractal dimension can be used to distinguish diabetic subjects from healthy subjects and hence this technique could be used in diagnosis of diabetes using images of retina. It is interesting that during other diagnostic procedures related to retina images, this information can be generated as additional information adding value to the diagnostic procedures. Details of implementation of the technique are presented.