A Clinical-Radiomics Model of Preoperative Prediction IMN Status of Breast Cancer Patients

Abstract

<h3>Purpose/Objective(s)</h3> To establish a prediction model based on clinicopathological factors and dynamic contrast enhanced (DCE)-MRI radiomics features, and to make a non-invasive prediction of IMN metastasis risk before operation. <h3>Materials/Methods</h3> The DCE-MR images and clinicopathological data of 124 breast cancer patients who had undergone internal mammary sentinel lymph node biopsy and / or internal mammary lymph node (IMN)dissection with IMN negative on preoperative images from January 2013 to December 2019 were obtained. All patients received plain breast scan and dynamic contrast-enhanced MR in Shandong Cancer Hospital before operation. In this study, three types of predictive models were established, which are based on clinicopathological risk factors, DCE-MRI imaging features and the combination of the two: 1) Clinical model: a logistic regression model based on clinicopathological factors was developed to predict ALN status. 2) Radiomics model: radiomics model based on radiomics score (RS) score is developed. 3) Clinical-radiomics model: combining the imaging features with the independent risk factors for independent prediction of IMN status, the multiple logistic regression method was used to establish a nomogram for predicting the risk of IMN metastasis. All models were evaluated by ROC and AUC, and clinical decision analysis curves was used to evaluate the clinical application value of the model. <h3>Results</h3> Three variables including primary tumor location, PR status and axillary lymph node metastasis status were included in the clinical model, and the AUC was 0.913. 150 imaging features were extracted from DCE-MRI images of primary tumors. Finally, 4 imaging features were selected to participate in the development of RS. Finally, the AUC of the imaging model combined with DCE-MRI imaging features of IMN was 0.831. The clinical-radiomics model with an AUC of 0.964. Delong test showed that, there was no significant difference in ROC curve between clinical model and radiomics model (p=0.1337) , but there were significant differences in ROC curve between clinical model and clinical-radiomics model, and between clinical model and clinical-radiomics model (p<0.05). DCA curve analysis shows that the application of clinical-radiomics model has the greatest net benefit, optimal predictive efficiency and better distinguishing ability, which is better than clinical model and radiomics model. <h3>Conclusion</h3> Clinical-radiomics model is helpful to evaluate the IMN metastasis status of breast cancer patients before operation.