Optimal Feature Selection and Automatic Classification of Abnormal Masses in Ultrasound Liver Images

Ultrasound imaging has found its own place in medical applications as an effective diagnostic tool. Ultrasonic diagnostics has made possible the detection of cysts, tumors or cancers in abdominal organs. In this paper, the possibilities of an automatic classification of ultrasonic liver images by optimal selection of texture features are explored. These features are used to classify these images into four classes-normal, cyst, benign and malignant masses. The texture features are extracted using the various statistical and signal processing methods. The automatic optimal feature selection process is based on the principal component analysis. This method extracts the principal features, or directions of maximum information from the data set. Using this new reduced feature set, the abnormalities are classified using the K-means clustering method. Based on the correct classification rate, a new optimal reduced feature set is created by combining the principal features extracted from the different texture features, to get a higher classification rate