MRI-based Radiomics For Preoperative Prediction Of Ki-67 Status In Breast Cancer

Background:Breast cancer is one of the most common malignancies amongwomen,accounting for 15%among all tumors in women.With the development of personalized precision medicine,tumor biomarkers have become increasingly important in clinical management.Among them,the expression level of Ki-67 plays an especially important role in the clinical diagnosis and treatment of breast cancer and can be used to assist clinical decision-making.Nowadays,the Ki-67 detection method of breast cancer is performed by immunohistochemistry(IHC).It is not only invasive and time consuming,but also costly.As a handy,non-invasive,and repeatable method,radiomics is expected to be a preoperative predictor of Ki-67 in breast cancer.Purpose:To investigate the value of a magnetic resonance imaging—based radiomics signature for preoperatively predicting the Ki-67 status in patients with breast cancer.Methods:This retrospective study has been approved by the Ethics Committee of Guangdong General Hospital(Guangdong Academy of Medical Sciences).Finally,318 patients with clinicopathologically confirmed breast cancer(training dataset:n=200;validation dataset:n=118)were included in the study.All patients underwent MR before treatment and had complete clinical and pathological datas.MR images were derived from the PACS system and regions of interest(ROI)were manually delineated by two experienced radiologists.Radiomics features were extracted from T2-weighted images(T2WI)and contrast-enhanced T1-weighted(T1+C)images of breast cancer.Matlab software was applied to extract radiomics features from the seleted ROI.The interclass correlation coefficient(ICC)was used to analyze the reproducibility of two observers in radiomics assessment of Ki-67 status in breast cancer.In the training dataset,features with ICCs>0.75 were included in the Least Absolute Shrinkage and Selection Operator(LASSO)regression model for feature selection and the radiomics signature buliding.The relationship between the radiomics signature and the Ki-67 status was assessed by Mann-Whitney U tests or independent samples t tests,where appropriate.The receiver operating characteristic curve(ROC)was firstly used to evaluate the predictive performances of the radiomics signature for Ki-67 status in the training dataset,and then validated in the validation dataset.Results:Ultimately,16 and 14 features based on T2WI and T1+C images,respectively,were selected to build the radiomics signatures.In the training and validation dataset,radiomics signatures based on T2WI were significant difference between the Ki-67 positive group and the Ki-67 negative group(both P<0.0001).The radiomics signature based on T1+C images was significantly correlated with the Ki-67 status in the training dataset(P<0.0001)but not in the validation dataset(P =0.083).The T2WI based radiomics signature exhibited good discrimination of Ki-67 status.In the training dataset,the AUC was 0.762[95%confidence interval(CI):0.685-0.838].The AUC of the validation group was 0.740(95%CI:0.645-0.836).Conclusion:Radiomics signature based on T2WI was an independent predictor of Ki-67 status in breast cancer,and has a high diagnostic efficiency.Radiomics signature serves as a non-invasive method that can be used to assist the preoperative prediction of Ki-67 status in breast cancer.