Automated differentiation of benign and malignant liver tumors by Ultrasound Images

Background & Aims: Early detection and reliable differentiation of benign and malignant liver tumors could lead to improved cure rate and costs. Ultrasound image (US) is a convenient medical imaging method for interpreting liver tumors. Visual inspection of ultrasound images sometimes is combined with error and needs biopsy to confirm whether a tumor would be benign or malignant. The aim of this study is to explore the potential of computerize texture analysis methods for classifying benign and malignant liver tumors in US imaging. Methods and materials: The US image database comprised 38 liver patients (25 malignant and 13 benign).Up to 270 texture features parameters as descriptors computed for each selected region of interest (ROIs) under default normalization scheme. Two feature reduction methods: Fisher and POE+ACC algorithms are applied to find the most effective features to differentiate benign from malignant liver. Obtained features parameters under two standardization states: standard (S) and nonstandard (NS) were used for texture analysis with PCA and LDA. Finally, Receiver Operating Characteristic (ROC) curve analysis was used via calculating sensitivity, specificity accuracy and Az value (area under the ROC curve) to examine the discrimination performance of applied texture analysis methods. Results: The very excellent performance for discrimination between benign and malignant liver tumors was recorded for LDA with sensitivity of 98.7%, specificity of 100% and Az value of 1. Also, for PCA discrimination results has sensitivity of 98.6%, specificity of 100% and Az value of 0.99. Conclusion: Our results indicates that texture analysis of the liver US images has potential to increase confidence of radiologist in classification of benign from malignant liver tumors.