Research On Diagnostic Models Of Benign And Malignant Breast Lesions Based On MR Radiomics

Objective: In this study,Magnetic Resonance(MR)radiomics were used to train multiple diagnostic models for the identification of benign and malignant breast nodules.Furthermore,unnecessary puncture biopsy or operation can be reduced by screening benign lesions.Methods:A total of 190 breast lesions,including 147 malignant lesions and 43 benign lesions,were collected from 180 patients who underwent plain and enhanced breast MR examination in the Department of Radiology,First Hospital of Jilin University from January 2019 to January 2022.All lesions were retrospectively analyzed.The opensource software 3D-slicer was used to manually delineate region of interest(ROI)on the fat suppression sequences of T2WI(T2-FS)and the first and second phase T1 WI enhancement sequences.RIAS software is used to convert DICOM format image information into NIFTI format and automatically extract radiomics features.All data were randomly divided into a cross-validation set(133 cases)and a validation set(57cases)according to a ratio of 7:3.In the cross-validation set,the radiomics features with good stability were initially screened out through intra-group and inter-group correlation analysis.Least absolute shrinkage and selection operator(LASSO)algorithm screened radiomics features signature related to the identification of benign and malignant breast lesions,and then used in the training model.Finally,three single sequence logistic regression(LR)models and three sequence fusion LR models were obtained by using the training set data.Finally,accuracy rate,receiver operator characteristic(ROC)curve area under the curve(AUC),sensitivity and specificity were uses to evaluate for the diagnostic efficacy of the model in the training set and the validation set,and the differences between different models were compared.Results:Among the four radiomics LR models established in this study,fusion model has the best diagnostic efficacy for benign and malignant breast lesions.And the AUC,sensitivity,specificity of the test set are 0.921,0.92,0.83 respectively,and are superior to the corresponding values of the other three single sequence models.Conclusion: Among the four kinds of radiomics models,the fusion model was the best in the diagnosis of benign and malignant breast lesions.Multi-sequence MRI fusion model can provide objective quantitative indicators for qualitative analysis of breast lesions and provide reference for whether needle biopsy or surgery should be performed.