MRI-based Radiomics For Prediction Of Ki-67 Changes After Neoadjuvant Chemotherapy In Breast Cancer

Background:Breast cancer ranks the first among female malignant tumors worldwide,and it is also the most commonly diagnosed cancer for incidence among women in China.Its growth rate has reached 3%-4%per year,significantly exceeding the global average level.Highly heterogeneity is well known in breast cancer,thus the efficacy of neoadjuvant chemotherapy(NAC)varies from diverse patients.Pathological complete response(pCR)is proposed as a predictive indicator for long-term benefit.However,it is vital to seek alternative markers for patients without achieving pCR.Ki-67 is recognized to be a molecular marker for predicting the rate of pCR,but the aspect is rarely covered in prediction for Ki-67 changes after neoadjuvant chemotherapy.As a noninvasive and reproducible technique,radiomics is expected to be a new method for predicting the changes in Ki-67.Purpose:To develop the value of radiomics signature based on magnetic resonance imaging in predicting changes in Ki-67 in breast cancer after neoadjuvant chemotherapy.Materials and Methods:This study was performed on breast cancer patients who underwent surgery after neoadjuvant chemotherapy from June 2014 to April 2018 in Guangdong Provincial People's Hospital.All patients had comprehensive clinical data and magnetic resonance examination before neoadjuvant chemotherapy,as well as the data about status of Ki-67 determined by immunohistochemistry of core biopsy sample and the surgical specimen.162 patients with positive expression of Ki-67 before neoadjuvant chemotherapy were included in the study,including 105 cases in training cohort and 57 cases in validation cohort.Tumors were manually segmented on T2-weighted images(T2WI)and contrast-enhanced T1-weighted images.Least absolute shrinkage and selection operator(LASSO)regression was performed to screen out the features with good agreement which were used for establishing radiomics signature and predictive model of changes in Ki-67 after adjuvant chemotherapy.Independent sample t test was used to compare and analyze the differences of radiomics signature in the changes of Ki-67 after NAC.The prediction performance of radiomics signature was evaluated by the receiver operating characteristic(ROC)curve for the changes of Ki-67 labeling index.Results:In the training cohort,the expression level of ki-67 after neoadjuvant chemotherapy turned to negative in 70 patients(positive group),and remained high level of expression in 35 patients(negative group).In the validation cohort,there are 46 and 11 cases of breast cancer in the positive and negative group respectively.16717 features were extracted from images of both T2WI and TIWI enhanced scan.16 radiomics features from T2WI and 2 radiomics features from T1+C were selected after dimension reduction to construct radiomics signature.The signature based on T2WI images showed a statistical significance between the positive group and the negative group(P<0.05 in both the training cohort and the validation cohort,0.000 and 0.045,respectively).Area under the curve(AUC)of both the training cohort and validation cohort in the prediction model is greater than 0.7,which is 0.877(95%Cl:0.803 to 0.951)and 0.711(95%CI:0.550 to 0.873),respectively.There was a significant difference in the signature between the positive group and the negative group(P=0.000)in the training cohort based on TIWI enhanced images,but there was no statistically significant in the validation cohort(P=0.597).Conclusion:The T2WI-based radiomics signature exhibited good performance for predicting a decrease in Ki-67 after NAC,which helps to predict prognosis and identify patients with high risk of recurrence and death.