A logistic regression model for predicting breast malignancy with ultrasonographic features

Objective To establish a logistic model for predicting breast malignancy on the basis of ultrasonographic features. Methods The features of gray-scale ultrasonography (US), color Doppler flow imaging (CDFI) and ultrasonic elastography (UE) were evaluated in 475 breast lesions confirmed by surgical pathology. A Logistic model for predicting breast malignancy on the basis of ultrasonographic features was obtained. A receiver operating characteristic (ROC) curve was used to assess the performance of the Logistic model. Results Seven ultrasonic features were finally entering the Logistic model, they were: elasticity score (odds ratio, OR=5.735), margin (OR=8.421), microcalcification (OR=8.755), color Doppler flow grade (OR=1.767),heterogeneous texture (OR=0.276), shadowing (OR=3.282), and enhanced transmission (OR=0.396). The area under the ROC curve was 0.979. Conclusion The Logistic model can help differentiate benign from malignant breast lesions.