ANN Classification of MS Subgroups with Diffusion Limited Aggregation

Karaca Y., Cattani C., KARABUDAK R.

18th International Conference on Computational Science and Its Applications (ICCSA), Melbourne, Australia, 2 - 05 July 2018, vol.10961, pp.121-136 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 10961
  • Doi Number: 10.1007/978-3-319-95165-2_9
  • City: Melbourne
  • Country: Australia
  • Page Numbers: pp.121-136
  • Hacettepe University Affiliated: Yes


Diffusion Limited Aggregation (DLA), the procedure in which substances blend irrevocably to produce dendrites, is idealised. During this process, the slowest phase tends to be the diffusion of substance to aggregate. This study focuses on the procedure where substances enduring a random walk because of Brownian motion cluster together to form aggregates of such particles. Magnetic Resonance Image (MRI) is one of the methods used for identifying nervous system chronic disorders. MS_ dataset, comprised of MR images belonging to patients with one of the MS subgroups, was used in this study. The study aims at identifying the homogenous and self-similar pixels that the regions with lesions are located by applying the DLA onto the patients' MR images in line with the following steps: (i) By applying the Diffusion Limited Aggregation (DLA) algorithm onto the MS_dataset (patients' MRI) the regions with the lesion have been identified. Thus, DLA_MS dataset has been generated. (ii) Feed Forward Back Propagation (FFBP) and Cascade Forward Back Propagation (CFBP) algorithms, two of the artificial neural network algorithms, have been applied to the MS_dataset and DLA_MS dataset. MS subgroups have been classified accordingly. (iii) Classification Accuracy rates as obtained from the application of FFBP and CFBP algorithms on the MS_dataset and DLA_MS dataset have been compared. Having been done for the first time, it has been revealed, through the application of ANN algorithms, how the most significant pixels are identified within the relevant dataset through DLA.