Image Processing of Finite Size Rat Retinal Ganglion Cells Using Multifractal and Local Connected Fractal Analysis

Submitted to the 17 Australian Joint Conference in Artificial Intelligence
Published in Lecture Notes in Computer Science, Springer-Verlag Heidelberg, Volume 3339 pp 961, November 2004.

Jelinek, H.F.,1, Cornforth, D.J.,2, Roberts, A.J.,3, Landini, G.,4, Bourke, P.D.,5 and Iorio, A.6

1 School of Community Health, Charles Sturt University, Australia
2 School of Environmental and Information Sciences, Charles Sturt University, Australia
3 Department of Mathematics and Computing, University of Southern Queensland, Australia
4 Oral Pathology Unit, School of Dentistry, University of Birmingham, U.K.
5 Astrophysics and Supercomputing, Swinburne Univ. of Tech., Australia
6 School of Computer Science and IT, RMIT University, Melbourne, Australia.


Automated image processing aids in classification of biological images. Fractal analysis has been used to characterize natural objects based on the assumption that these objects represent monofractals. However, many natural structures such as neurons may not belong to this category. The multifractal spectrum may mitigate this difficulty, provided that the question whether neurons are multifractal is satisfactorily answered. Here we report the outcome of applying three methods that elucidate the variation within 16 rat retinal ganglion cells as an example of a neurons using the local connected fractal dimension (LCFD), mass-radius (MR) and maximum likelihood multifractal (MLMF) analyses. Our results based on LCFD indicate that the neurons studied are possibly multifractal. However utilizing the MR method provided inconclusive results due to the finite size of the cells and the density variation throughout their structure. This has been addressed by utilizing a novel unbiased method - the MLMF method.