A reliability study of fractal analysis of the skeletonised vascular network using the "box-counting" technique

Binary vascular networks of the human retina were obtained by computerized processing of digital fundus images. Fractal analysis was performed on skeletonised versions. The effect of three parameters in the vascular segmentation and skeletonization algorithm was quantified, with threshold level found to have the greatest influence. The results were compared to fractal analysis of skeletons derived by manual tracing. The mean difference in fractal dimension between 2 observers is 0.004 (0.3%) and the coefficient of repeatability is +/- 0.050 (3.4%). Between a single observer and the computerized approach the mean difference is 0.012 (0.8%) and the coefficient is +/- 0.038 (2.6%). The computerized approach demonstrated a superior reliability compared to manual segmentation.