Theory, Analyses and Predictions of Multifractal Formalism and Multifractal Modelling for Stroke Subtypes’ Classification

Abstract

Fractal and multifractal analysis interplay within complementary methodology is of pivotal importance in ubiquitously natural and man-made systems. Since the brain as a complex system operates on multitude of scales, the characterization of its dynamics through detection of self-similarity and regularity presents certain challenges. One framework to dig into complex dynamics and structure is to use intricate properties of multifractals. Morphological and functional points of view guide the analysis of the central nervous system (CNS). The former focuses on the fractal and self-similar geometry at various levels of analysis ranging from one single cell to complicated networks of cells. The latter point of view is defined by a hierarchical organization where self-similar elements are embedded within one another. Stroke is a CNS disorder that occurs via a complex network of vessels and arteries. Considering this profound complexity, the principal aim of this study is to develop a complementary methodology to enable the detection of subtle details concerning stroke which may easily be overlooked during the regular treatment procedures. In the proposed method of our study, multifractal regularization method has been employed for singularity analysis to extract the hidden patterns in stroke dataset with two different approaches. As the first approach, decision tree, Naive bayes, kNN and MLP algorithms were applied to the stroke dataset. The second approach is made up of two stages: i) multifractal regularization (kulback normalization) method was applied to the stroke dataset and mFr_stroke dataset was generated. ii) the four algorithms stated above were applied to the mFr_stroke dataset. When we compared the experimental results obtained from the stroke dataset and mFr_stroke dataset based on accuracy (specificity, sensitivity, precision, F1-score and Matthews Correlation Coefficient), it was revealed that mFr_stroke dataset achieved higher accuracy rates. Our novel proposed approach can serve for the understanding and taking under control the transient features of stroke. Notably, the study has revealed the reliability, applicability and high accuracy via the methods proposed. Thus, the integrated method has revealed the significance of fractal patterns and accurate prediction of diseases in diagnostic and other critical-decision making processes in related fields.