Characterization of liver tissues in ultrasound images using texture analysis

Abstract

This study is an attempt to study the liver ultrasound images using computer analysis techniques and hence the main objective of this study was to characterize the liver tissues in an ultrasound images into three classes which includes; fatty, cirrhosis and normal tissue types by using texture analysis. The texture were extracted from spatial gray level dependence matrix using a window of 20×20 pixels of angle zero and distance equal one pixel. The images were collected from 60 patients represents the classes of the study in the period from 6/2011 to 2/2012. The images were scored by an expert two sonologist where the scoring was accepted in case of agreement between the two of them. Then the textural features were extracted from selected sub-images that show only the class of interest. The classification technique were adopted as a method of pattern identification the images into three classes. A linear discriminant analysis using stepwise were used to classify the sample into the predefined classes. The stepwise selected 9 features out of fifteen features as the most discriminant features; they included: sum variance, Entropy, Energy, Inverse difference moment, Correlation, sum entropy, Difference average, Information1 and mean. The result of this study showed that the total classification accuracy was 93.3%, with an accuracy of 85.1% 98.4% and 94.9% for fatty, cirrhosis and normal tissue respectively.