Computer-Aided Diagnosis Approaches to Fatty Liver Disease According to Sonographic Images Based on Wavelet Transform: A Review Study

Abstract

Introduction: Fatty liver is usually diagnosed by ultrasound, but this diagnosis can be difficult because the disease does not always lead to abnormal conditions on gray levels that can be detected by the eye. However, ultrasound is still the first choice to detect fatty liver due to its low cost and availability, and the lack of side effects. The study reviewed Computer-Aided Diagnosis approaches to fatty liver disease, based on wavelet transform sonographic image processing. Methods: In this review study, a search was conducted based on related keywords and articles that had been published in English over the last 12 years. The findings were extracted based on the aim of study. Findings: Nowadays wavelet transformation has been widely used in the field of medical image processing because of its adaptability to the characteristics of the human eye system. The well-known wavelets used to liver diseases detection include Haar, Symlet, Daubechies and Gabor. Extracting the proper properties of images plays an important role in detecting diseases. Important statistical features of image textures are: statistical descriptors based on the intensity histogram and the GLCM matrix (Gray level Co-occurrence Matrix). The popular algorithms used for liver disease include neural network, Support Vector Machine (SVM), Bayesian, decision tree, K-Nearest Neighbor (KNN), and regression. Conclusion: The sensitivity, specificity and accuracy of the extracted statistical features of the output components of wavelet transform are generally better than those obtained from the original image itself. Gabor’s wavelet transformation often has a higher efficiency than the Daubechies and Symlet wavelet transforms because the two transforms only break up the halfband of low frequencies and lose some of the intermediate frequency regions, while Gabor retains all of the frequency regions This precision also mainly depends on the type of features selected and the type of classification. Statistical features based on intensity histograms do not provide relative information about the spatial of pixels relative to each other. To enter this spatial information of pixels in a texture analysis, it is recommended to use GLCM matrix in gray images. The type of classifier used can significantly impact on the precision of the final diagnosis.