Computer-aided analysis of ultrasound elasticity images for classification of benign and malignant breast masses.

OBJECTIVE

The purpose of this study was to evaluate computer-aided analysis of ultrasound elasticity images for the classification of benign and malignant breast tumors.

MATERIALS AND METHODS

Real-time ultrasound elastography of 140 women (mean age, 46 years; age range, 35-67 years) with nonpalpable breast masses (101 benign and 39 malignant lesions) was performed before needle biopsy. A region of interest (ROI) was drawn around the margin of the mass, and a score for each pixel was assigned; scores ranged from 0 for the greatest strain to 255 for no strain. The diagnostic performances of a neural network based on the values of the six elasticity features were compared with visual assessment of elasticity images and BI-RADS assessment using B-mode images.

RESULTS

The values for the area under the receiver operating characteristic curve (A(z)) of the six elasticity features--mean hue histogram value, skewness, kurtosis, difference histogram variation, edge density, and run length--were 0.84, 0.69, 0.63, 0.75, 0.68, and 0.71, respectively. The sensitivity, specificity, positive predictive value, and negative predictive value of the neural network based on all six features were 92% (36/39), 74% (75/101), 58% (36/62), and 96% (75/78), respectively, with an A(z) value of 0.89, which is significantly higher than the A(z) of 0.81 for visual assessment by radiologists (p = 0.01) and 0.76 for BI-RADS assessment using B-mode images (p = 0.002).

CONCLUSION

Computer-aided analysis of ultrasound elasticity images has the potential to aid in the classification of benign and malignant breast tumors.