A Machine-Learning Model for the Prediction of Triple-Negative Breast Cancer Based on Multiparameter MRI.

Abstract

Objective: To explore the difference between triple-negative breast cancer (TNBC) and non-triple-negative breast cancer (non-TNBC) based on multi-parametric MRI imaging features and construct a prediction model to identify TNBC.

Methods: A retrospective study enrolled 1353 women with 1376 malignant lesions who had no additional therapy before surgery between January 2019 and December 2020 in a single center. The images were accessed according to BI-RADS-MR^® (fifth ed.) atlas. The lesions were classified as TNBC group and non-TNBC and then randomly divided into a primary cohort (n = 963) and a validation cohort (n = 413) at a ratio of 7:3. In the primary cohort, univariate analysis, logistic regression analysis and Boruta algorithm were used to determine the independent predictors for TNBC and non-TNBC. The machine learning classifier XGboost was developed based on the features to predict TNBC. The area under the receiver operating characteristic (ROC) curve (AUC) was applied to evaluate the model prediction ability. The diagnostic performances of the model were evaluated in the validation cohort.

Results: Necrosis, edema, the maximum diameter of lesions, enhancement ratio in each phase, time to peak, gland enhancement ratio, wash-in slope and the number and diameter of the vessels were independent predictors predicting TNBC. The AUCs of the model were 0.795 (95% CI: 0.758-0.832) and 0.705 (95% CI: 0.640-0.770) in the primary cohort and validation cohort, respectively.

Conclusion: The model based on multiparameter MRI has good predictive ability and can be used to predict the probability of TNBC.