Value of prediction model for BI-RADS imaging features in nonpalpable calcified breast lesions / 实用放射学杂志

ABSTRACT

Objective To improve the diagnostic accuracy of nonpalpable calcified breast lesions by establishing a Logistic multivariate prediction model to assess the probability of benign/malignant breast lesions.The proposed model is based on the clinical and BI-RADS-X-ray imaging features of patients with nonpalpable calcified breast lesions.Methods A total of 147 nonpalpable calcified breast lesions were analyzed retrospectively.Firstly, based on the personal experience,the X-ray imaging data of lesions were analyzed to obtain the BI-RADS categorization, and the ROC curve was plotted by comparison with pathology.Then the univariate and multivariate analysis was performed on the clinical and X-ray imaging features of pathology to select the independent factors related to benign/malignant features.Further,a Logistic regression model was built,the suitable cut-off point was determined, and the ROC curve was obtained.Finally,the comparisons of the diagnostic accuracy of breast lesions were made between the method using the BI-RADS categorization and the method using the Logistic regression model.Results The AUC of the BI-RADS method was 0.867 9.The univariate analysis showed that there exist statistical differences among clinical features of patients(age,location,and quadrant),as well as the BI-RADS-X-ray imaging features (distribution,morphological and gland density).Also,by using the multivariate Logistic regression equation,the statistical differences among age,quadrant and morphological difference can be observed.The AUC using the built Logistic regression model was 0.906 3.Conclusion The diagnostic accuracy of breast lesions using the Logistic model is higher than that using the BI-RADS categorization method.Therefore, the proposed model is valuable for obtaining accurate diagnosis of breast lesions.