3.0 T multi-parametric MRI combined with clinical features improve malignancy prediction of BI-RADS 4 lesions and preoperative prediction of Nottingham Prognostic Index.

PURPOSE

To establish an optimal model to improve the malignancy prediction of BI-RADS 4 lesions and the preoperative prediction of tumor prognosis.

MATERIALS AND METHODS

Ninety-six patients with 126 histopathology-confirmed breast lesions were included in the study. Conventional imaging features, radiomic features based on 3.0 T multi-parametric MRI and patient`s clinical characteristics were analyzed and selected as model candidate features. The least absolute shrinkage and selection operator (Lasso) and Random Forest (RF) were used to construct the combined model. Receiver operating characteristic (ROC) and Net Reclassification Improvement Index (NRI) were performed to assess the diagnostic efficiency between the model and BI-RADS category. Relative ratio (RR) was calculated to assess the ability of model to predict the invasiveness of breast cancers. Finally, the malignant probability (MP) calculated by the optimal model, MRI-based size and lymph node (LN) stage were used by logistic algorithm to construct a preoperative Nottingham Prognostic Index (NPI) model.

RESULTS

The combined model incorporating multi-parametric MRI and clinical characteristics was superior to BI-RADS category in the diagnosis of breast cancer (NRI: 1.71, p < 0.05), and had an accuracy of 94 % to predict the malignancy of BI-RADS 4 lesions In addition, MP calculated by the combined model in association with MRI-based size and LN stage can accurately predict the NPI preoperatively (AUC: 92.1 %).

CONCLUSIONS

The combined model based on multi-parametric MRI and clinical characteristics improves the malignancy prediction of BI-RADS 4 lesions and the preoperative prediction of NPI, therefore providing comprehensive information on the characteristics and treatment plans for breast cancer.