New method for texture classification based on local surface structure and its application to ultrasonic liver images

In this paper, the classification of ultrasonic liver images is presented. By combining texture features extracted by two methods, including co-occurrence matrix and local texture structure frequency, the system can classify three liver states, which are normal liver, hepatitis and cirrhosis. The co-occurrence method extracts texture features which represent gray level variation of pixel pair under a specific spatial relationship. The local texture structure frequency, we have proposed new texture descriptor, extract local surface structure of a small texture unit. Several texture features are derived from this texture frequency, which include coarseness, homogeneity, structure stability, etc., A forward sequential search process is adopted to look for the most useful texture features from co-occurrence matrix and our proposed method. These textures features extracted from the two methods are then fed into a probabilistic neural network to classify the liver states. The system has been implemented on a Sun sparc II workstation and tested with sixty ultrasonic images. The correct classification rate is around 88%. These results suggest that the selected texture features are effective for the classification of liver echotexture.