Development and validation of a model for predicting malignant probability of asymmetries signs of BI-RADS category 4 or 5 in mammography / 中华放射学杂志

ABSTRACT

Objective:To develop a model for predicting malignant probability of asymmetries signs of BI-RADS category 4 or 5 in mammography, and to evaluate its predictive performance.Methods:The clinical and imaging data of 187 patients with asymmetrical signs of BI-RADS 4 or 5 who underwent mammography in Xinhua Hospital Affiliated of Shanghai Jiao Tong University School of Medicine from January 2014 to September 2017 were analyzed retrospectively. According to the examination time, 187 patients were divided into training set (109 cases) and verification set (78 cases). The clinical and mammographic features of the patients were analyzed. The predictors were screened by univariate and multivariate logistic regression analysis. Then a predictive model was constructed based on the results with a nomogram drawn. Performances of predictive models were evaluated with area under the curve (AUC) of ROC,calibration curve,and decision curve analysis (DCA).Results:A total of 187 asymmetries signs including 72 malignant and 115 benign was collected. Six predictive factors were harvested to construct the predictive model, which included menstrual status, the number of visible standard view of asymmetries signs, whether asymmetries signs locating at the peripheral layer of tissue, whether asymmetries signs associating with suspicious calcification, skin thickening, or nipple retraction. The predictive model had good performance in both training set and verification set, with the AUC values of 0.85 and 0.84, respectively. The DCA curve showed that the nomogram had good clinical application values.Conclusion:The predictive model can predict malignancy probability of BI-RADS category 4 or 5 asymmetries, which can be used as a referable method for radiologist to evaluate mammographic asymmetries signs.