Multifractal Analysis with L2 Norm Denoising Technique: Modelling of MS Subgroups Classification

Abstract

Multifractal methods are employed to recognize the significant attributes through the removal of the noise in any dataset. Being a useful and sensitive technological tool, magnetic resonance imaging (MRI) is employed for the diagnosis of chronic diseases of the central nervous system, one of which is Multiple Sclerosis disease (MS). MS subgroups (Primary Progressive MS (PPMS), Relapsing Remitting MS (RRMS) and Secondary Progressive MS (SPMS)) were examined in this study. The dataset is based on the Expanded Disability Status Scale (EDSS) and MRI that belong to 139 individuals in total, 120 of whom have one of the MS subgroups, and 19 are healthy individuals. The stages of the study are as follows: (i) L2 Norm Denoising, which is a multifractal method was applied on the MS dataset, namely the MS dataset. From this application, the new dataset was formed. (ii) The new dataset (MS_L2 Norm dataset) which is comprised of the significant attributes was obtained as a result of the application in (i). (iii) K-means clustering algorithm was applied on both the MS dataset and MS_L2 Norm dataset for the classification of the MS subgroups. (iv) The classification accuracy rate for the K-means clustering algorithm in relation to the MS dataset and MS_L2 Norm dataset was compared both for patients with MS subgroups and healthy individuals. For the first time, through this study, the K-means classification has been performed by using multifractal denoising methods for the aforementioned dataset. The results of our study show that K-means algorithm and classification yielded more accurate results for the MS_L2 Norm dataset compared to the MS dataset in terms of classification. Overall, this study has aimed to bridge a gap in the literature.