Comparative analysis of texture characteristics of malignant and benign tumors in breast ultrasonograms.

Abstract

We evaluated various texture features and region of interest (ROI) types of breast ultrasonograms in order to determine the best-performing combinations for differentiating between benign and malignant solid breast nodules. A total of 21 breast ultrasonograms (12 benign, nine malignant) containing solid breast nodules were evaluated. Eight ROI types were defined around the nodules. The texture feature of each ROI was measured and the ratios of texture features were calculated for each pair of ROIs. This procedure was repeated for five different feature types, thus yielding texture feature ratios for 140 different combinations of ROIs and texture features. We evaluated the performance of the texture feature ratio in differentiating between benign and malignant nodules using t test analysis. Evaluating the top ranked texture and ROI combinations, we found edge density and mutual information were the best two texture features, and that the ROI types of outside lesion and lesion margin had good performance.