Differentiation of solid benign and malignant breast masses by quantitative analysis of ultrasound images

Although breast sonography is highly accurate at distinguishing solid from cystic lesions, it is less precise when differentiating benign and malignant masses. The goal of this study is to evaluate quantitative methods for differential diagnosis of solid breast masses. Three margin features extracted from B-Mode images, along with age of the patient, were analyzed by logistic regression to classify lesions as malignant or benign. Receiver-operating characteristic (ROC) analysis assessed the diagnostic performance of each feature individually as well as collectively. In each case, classification performance was evaluated using a leave-one-out cross validation method. The robustness of the diagnostic algorithm was determined by studying effects of sample size and training level on the overall performance of the classifier. The results show that while there was no statistical difference in the size of benign and malignant lesions, the margin characteristics of the masses varied significantly, differentiating the two groups. When all the features were used together, the probability of malignancy for the solid breast masses could be determined with ROC area ranging from 0.83 to 0.89. Diagnostic accuracy was notably influenced by the size of the database, the ratio of malignant to benign lesions, and the training level of the algorithm. Significant diagnostic performance can be attained using margin characteristics of the lesions and logistic regression classifiers.