Prediction Of Treatment Response To Neoadjuvant Therapy In Breast Cancer With MRI And Staging CT-based Radiomics

BackgroundNeoadjuvant therapy is commonly used for clinically axillary lymph node-positive(cN+)breast cancer,which can help to reduce tumor burden preoperatively and increase the likelihood of conservation surgery for the breast and axilla Treatment response to neoadjuvant therapy is an important prognostic predictor in breast cancer,and patients with pathologic complete response(pCR)would have favorable outcome.In addition,treatment response can also guide the subsequent surgery plan.Therefore,it is of great clinical significance to accurately predict treatment response to neoadjuvant therapy before surgery.Purpose1)To investigate the clinical application value of the combination of MRI qualitative features and clinicopathologic features for preoperative prediction of axillary pCR to neoadjuvant therapy in breast cancer;2)Based on the research findings in the Part 1,to further investigate whether the quantitative MRI-based radiomics features could improve the accuracy of axillary pCR prediction;3)To explore the added value of pretreatment staging CT for prediction of treatment response to neoadjuvant therapy in breast cancer with radiomics approach.Methods1)A total of 389 patients with cN+breast cancer who received neoadjuvant therapy were retrospectively collected from 3 hospitals,and were chronologically divided into a training cohort(n=183),internal and external validation cohorts(n=76;n=75),and internal and external test cohorts(n=30;n=25).In the training cohort,univariate and multivariable logistic regression analyses were used to develop a simplified scoring system based on MRI and clinicopathologic features.In the validation cohort,the receiver operating characteristic(ROC)curve and calibration curve were applied to assess its predictive performance,which were further confirmed in the test cohorts.2)Study population were the same as the training cohort(n=183)and internal validation cohort(n=76)in the Part 1.Based on breast MRI,radiomics features were extracted.After feature selection,we used backpropagation neural network to construct a radiomics signature.Then,combined with the simplified scoring system of the part 1,the radiomic combined model was further developed,and its prediction performance was evaluated with ROC curve,calibration curve and decision curve analysis.3)A total of 215 breast cancer patients undergoing chest contrast-enhanced CT examination before neoadjuvant therapy were retrospectively enrolled from our hospital,and were divided by time into a training dataset(n=138)and a validation dataset(n=77).Radiomics features were extracted from the intratumoral and peritumoral regions of CT images,and after feature selection,we used logistic regression to construct intratumoral and peritumoral signature,respectively.A radiomic nomogram was further developed by incorporating the signatures with molecular characterization.Its predictive performance was evaluated in terms of discrimination,calibration and clinical utility.Results1)The simplified scoring system consisting of 7 MRI and clinicopathologic features,demonstrated good calibration and discrimination for pCR prediction,with AUCs of0.835,0.828 and 0.798 in the training cohort,internal and external validation cohorts,respectively.The axillary pCR rates were increased with the score,and patients with a score≥11 points had a pCR rate of 86%-100%.In the test cohorts,the diagnostic accuracy for axillary pCR was 80%-90%.2)The MRI-based radiomics signature including 5 features,achieved an AUC of 0.821 in the validation cohort,along with good calibration.The AUC of the radiomic combined model was 0.912 in both the training cohort and the validation cohort,and its predictive ability was superior than the simplified scoring system.The decision curve analysis indicated good clinical utility of the combined model.3)Compared to the clinical model constructed by molecular characterization alone(AUC:0.717;0.756),the CT-based radiomic nomogram had better predictive performance,with AUCsof0.821 and 0.818 in the training and validation dataset,respectively.The nomogram also had good calibration,and the decision curve analysis showed it was clinically useful.ConclusionWith qualitative and quantitative imaging features,this study can provide complementary tools for accurate prediction of treatment response to neoadjuvant therapy in breast cancer,and thus assist clinical individualized management.