Characterization of Liver Disease using Image Texture Analysis

This study aimed to use the texture analysis and classification methods to characterize the normal liver and liver pathologies in US images using image processing program (IDL, interactive data language). A prospective study for 50 patient with normal liver and same number of patient with hepatitis B, C and liver cirrhosis was examined using ultrasound scan after successful laboratory investigation and diagnostic confirmation also. Ultrasound images was acquired using mobile GE LOGIQ-5, US scanner and the image extracted and treated for image processing analysis using interactive data language. Tiff format was created as IDL variables and then using 3x3 window the image was scanned and based on the image histogram the selected feature also called FOS was calculated using this window. Linear discriminant analysis was used for the tissue classification. The study found that the liver cirrhosis, hepatitis (B), and (C) texture reveal a different underlying pattern compared to the normal liver tissues with classification sensitivity and specificity 97.1 %, 69.8 %, 73.3 %, and 92.5 % respectively, and the combination of the texture features throughout the different ultrasound images provide the highest predictive overall accuracy of 84.9 % using linear discriminant analysis.