Liver Pathological States Identification with Self-organization Models Based on Ultrasound Images Texture Features

Abstract

The article deals with the new possibilities in the development of diagnostic decision support systems on the example of normal and pathology states separation in diffuse liver diseases based on statistical features of ultrasound image texture. It is proposed to use the difference conversions of the original grayscale matric, which makes the results of image processing independent of variations in the original grayscale settings. Based on grey-level co-occurrence matrices, it was proposed to calculate a number of new indicators, which allowed to effectively distinguish the class textures of liver norm and pathology. The indicators' definitions are protected by a patent of Ukraine. The liver state classifiers are built using the GMDH Shell DS software in the form of analytical expressions and an authors' algorithm in the form of a forest, whose trees are built according to the Group Method of Data Handling principles. The work was performed on data, had provided by the Nuclear Medicine and Diagnostic Radiology Institute of the National Academy of Medical Sciences of Ukraine.