| Abstrakt: |
To develop an interpretable machine learning model based on DCE-MRI semi-quantitative parameters and clinical features to predict luminal and non-Luminal breast cancer. A total of 196 patients with breast MRI examination in our hospital from May 2021 to April 2023 were retrospectively enrolled, including 6 patients with bilateral breast cancer, and a total of 202 lesions were included. The patients were randomly divided into the training set (n =141) and the test set (n =61) at a ratio of 7:3. The enhanced lesions were delineated for post-processing, and the time-intensity curve(TIC) were obtained for semi-quantitative perfusion analysis. The clinicopathological and imaging features of the patients were collected according to the BI-RADS lexicon. Semi-quantitative parameters and clinicopathological and imaging features were screened by univariate analysis.The XGboost classifier was used to construct the model, and the area under the receiver operating characteristic (ROC) curve (AUC), accuracy, sensitivity, and specificity were used to evaluate the performance of the model. Shapley Additive explain (SHAP) technique was used to evaluate the global and local influence of predictors on the model output to explain the model. Compared with non-Luminal breast cancer, luminal breast cancer had lower washout and ADC values and higher SER. Patients with low HER-2 expression, smaller maximum tumor diameter and no tumor necrosis were more likely to be expressed as luminal breast cancer. The interpretable machine learning model based on the above features had an AUC of 0.842, accuracy, 0.680; sensitivity, 0.944; specificity,0.360; and F1, 0.791. Interpretable machine learning models based on DCE-MRI semi-quantitative parameters and clinical features can effectively predict luminal and non-Luminal breast cancer and provide a basis for the treatment decision-making of early breast cancer patients. [ABSTRACT FROM AUTHOR] |
