Model Construction Of Multimodal Imaging BI-RADS For Predicting Malignant Risk Of Breast Tumors

Objective: Combined with common clinical imaging examination methods,the superior indicators of ultrasound,mammography and magnetic resonance were combined to construct a comprehensive imaging BI-RADS risk prediction model suitable for breast mass and non-mass lesions.Methods: Patients who underwent ultrasound,mammography and magnetic resonance imaging before surgery in Shanghai Ruijin Hospital from August 2019 to September2020 were collected.Each patient selected one of the most suspicious lesions for evaluation,including 1717 mass lesions and 350 non-mass lesions.In view of the slightly different evaluation indicators of lesion types(mass/non-mass),the author made a joint diagnosis on the premise of mass and non-mass at the beginning.The lesions were clearly defined: that is,a mass that was clearly visible on an examination image was defined as a mass lesion,otherwise it was a non-mass lesions.The imaging features of the lesions were described and recorded according to the BI-RADS lexion of the three images,and the clinical factors of the patients were included.univariate factor logistic regression analysis was performed for mass lesions,and the indicators with statistically significant differences were further subjected to multivariate logistic regression to obtain the BI-RADS categories predicted by multiple combined diagnostic models.The diagnostic efficiency between models was compared by ROC curve.Secondly,for non-mass lesions,LASSO-Logistic regression algorithm was used to screen out indicators that can identify benign and malignant lesions.Nomogram was constructed with the selected indicators,and the probability of malignant tumors in each patient was obtained.Non-mass lesions were reclassified according to the positive predictive value of the BI-RADS lexion.Results: The positive predictive value of multiple models for the prediction of BIRADS categories for mass lesions:(1)MRI+Mammography+US: 1.96% for category3,9.32% for category 4A,41.24% for category 4B,85.04% for category 4C,96.83%for category 5;(2)MRI+Mammography: 0.00% for category 2,2.68% for category 3,12.40% for category 4A,47.09% for category 4B,88.20% for category 4C,97.77% for category 5;(3)MRI+ US: 0.00% for category 3,8.87% for category 4A,38.54% for category 4B,86.09% for category 4C,96.63% for category 5;(4)Mammography+ US:7.69% for category 3,15.38% for category 4A,59.60% for category 4B,87.21% for category 4C,97.12% for category 5.The positive predictive value of the non-mass lesion model for predicting BI-RADS category: 0.00% for category 2,0.00% for category 3,6.25% for category 4A,26.16% for category 4B,80.84% for category 4C,and 97.33% for category 5.Conclusion: The positive predictive values of the integrated image BI-RADS categories predicted by the model are all within the reference range of the guide.It makes the diagnosis and selection of reasonable treatment methods for breast mass and non-mass lesions are unified,which has high clinical application value.