Wavelet-based Multifractal Spectrum Estimation in Hepatitis Virus Classification Models by Using Artificial Neural Network Approach

Abstract

Fractal and multifractal geometries have been applied extensively in various medical signals which exhibit fractal characteristics. Application of such geometries rests on the estimation of fractal features. Within this framework, various methods have been proposed for the estimation of the multifractal spectral or fractal dimension of a particular signal. Wavelet transform modulus maxima (WTMM) is one of the methods employed for the detection of fractal dimension of a signal. It was developed for the characterization of signal singularities. Hepatitis, inflammation of the liver, may prove to be a serious disease with serious potential risks. This study proposes an alternative method for the classification of hepatitis virus as per die/live with the use of two aspects, namely multifractal analysis and Artificial Neural Network (ANN). As the first aspect, for the multifractal analysis, Wavelet Transform Modulus Maxima (WTMM) (Multifractal Spectrum estimation) was used with the following stages: (a) WTMM was applied to the hepatitis dataset (self-similar and significant attributes were identified) and wtmm_hepatitis dataset was generated. (b) Continuous Wavelet Transform was applied on the hepatitis dataset (hepatitis_dataset) and wtmm_hepatitis dataset. The second aspect is related to the application of Feed Forward Back Propagation (FFBP) algorithm which is an ANN application with the following steps: (i) FFBP algorithm was applied to both hepatitis dataset (hepatitis_dataset) and (wtmm_hepatitis dataset) to identify the classification as per die/live (ii) The attributes proven to be the most effective were determined based on the results (sensitivity, specificity and accuracy rate). The highest level of accuracy has been obtained from the wtmm_hepatitis dataset. The main contribution of this study is that it has proven to provide an alternative in multifractal spectrum estimation by determining the self-similar and significant attributes through WTMM for the first time in the literature. The proposed method in the study aims at bringing a new frontier in the related fields by placing emphasis on the significance of significant attributes’ characterization to obtain optimal accuracy rates for the solution of problems.