Enhancement of textural features in normal & diseased ultra sonogram of liver by Gaussian smoothing

Texture analysis is a very powerful tool to analyze a given image. Texture analysis based on spatial gray-level dependence (SGLD) matrix computation was carried out over the input ultrasound images of livers to extract features, namely, maximum probability, uniformity, entropy, element difference moment of order 2, inverse element difference moment of order 2, homogeneity, correlation and contrast. In this work seventy-six ultrasound scan images of human normal livers and twelve ultrasound images of fatty livers as identified by the physician were collected from hospital. From each filtered image (by Gaussian filter) portion of the liver five sub-images were suitably cropped and taken as an input image for analysis. An unsupervised neural network learning technique, namely, Self Organizing Map (SOM) was employed to examine various cluster formations based on certain combination of textural features extracted from normal as well as fatty livers. The findings establish a fact that textural features in normal and diseased ultra sonogram of liver have been enhanced by Gaussian smoothing.