Development And Validation Of A Magnetic Resonance Imaging Radiomics-based Model For Prediction Of Distant Metastasis Before Treatment Of Breast Cancer

Objective:We aimed to identify a magnetic resonance imaging-based model for assessment of the risk of individual distant metastasis(DM)before treatment of breast cancer(BC)Materials and Methods:The institutional review board of Shunde Hospital approved the retrospective analysis of data and the requirement for informed consent was dropped.The retrospective cohort analysis included 154 breast cancer patients.We acquired axial contrast-enhanced T1-weighted and T2-weighted images.Segmentation for regions of interst(ROI)was performed using 3D Slicer software.We extracted the radiomics features of intratumoural region,peritumoral region,intratumoural and peritumoral region.All patients were randomly divided into a training cohort(n=106)and a validation cohort(n=48)in a ratio of 7:3.We used univariate analysis and multivariate analysis to select the features.Using the features,we construted seven models for predicting distant metastasis of breast cancer,including a model of breast cancer radiomics features,a model of peritumoral radiomics features,a model combining breast cancer with peritumoral radiomics,a model combining breast cancer radiomics,clinical and pathological features,a model combining peritumoral radiomics,clinical and pathological features,a model combining breast cancer radiomics,peritumoral radiomics,clinical and pathological features and a model combining clinical and pathological features.We construted a model combining radiomics features,clinical and pathological features,with coefficients weighted by logistic regression analysis in the training cohort.An optimal cutoff value for classifying the patients into low-risk and high-risk groups based on the risk of DM.Kaplan-Meier survival curves were used to compare 3-year survival between the low-risk and high-risk groups.Results:Among seven models,the model of breast cancer radiomics features and the model combining radiomics features,Clinical and pathological features Showed a significant predictive ability.The AUC of the two models in the training cohorts are 0.886 and 0.902,and in the validation cohorts are 0.850 and 0.830.The model of breast cancer radiomics features and the model combining radiomics features,Clinical and pathological features better than a model combining clinical and pathological features.The three model combining radiomics features,Clinical and pathological features and the model combining Clinical and pathological features all comprised TNM stage and Ki 67.The model of breast cancer radiomics features and the model combining radiomics features,Clinical and pathological features had the same radiomics features.We construted a model retaining eight radiomics features and two clinical features finally,with coefficients weighted by logistic regression analysis in the training cohort.An optimal cutoff value for classifying the patients into low-risk and high-risk groups based on the risk of DM.Kaplan-Meier survival curves showed that the 3-year survival rate of breast cancer patients in high-risk group was significantly lower than that low-risk group(P<0.001).Conclusions:TNM stage,Ki67 and radiomics features are obviously related to distant metastasis of breast cancer.We construted a model combining radiomics features,Clinical and pathological features for the prediction of distant metastasis before initial treatment in patients with breast cancer.The nomogram as a tool that can help clinicians in identifying patients with a high risk of distant metastasis and optimizing therapeutic strategies.