Automatic Classification of Benign And Malignant Liver Tumors In Ultrasound Images

Introduction: Differentiation of benign and malignant liver tumors is very important for finding appropriate treatment procedure. Human eyes sometime are not able to diagnose the type of liver tumor. Texture analysis is considered as a suitable method to increase the diagnostic power of medical images. In this study texture analysis is employed in order to classification of benign and malignant liver tumors in ultrasound images. Materials and Methods: Our study was performed on 38 patients (25 malignant and 13benign). After selection of proper ROI, about 270 parameters were extracted from each ROI. Fisher algorithm was employed for selection of the best discriminating features. Three analysis methods (PCA, LDA and NDA) were performed at the last stage. Statistical analysis of data was performed by Receiver Operating Characteristic (ROC).   Results: The result of our study was very plausive. The best discrimination result was belong to NDA with sensitivity and specificity of 100%. PCA and LDA also had good results (sensitivity of 97% and 98% respectively).   Conclusion: This study show that texture analysis is able to classify benign and malignant liver tumors in US images with the high performance quality. Therefore, texture analysis may be a useful method for classification of benign and malignant liver tumors.