Prediction Of Pathological Complete Response To Neoadjuvant Chemotherapy In Breast Cancer Based On MRI Intratumoral And Peritumoral Radiomics

Objective: To investigate the predictive value of intratumoral and peritumoral MRI radiomics for pathological complete response(pCR)of Neoadjuvant chemotherapy(NAC).Methods: In this study,94 patients who were diagnosed with breast cancer through surgery or biopsy pathology and underwent pre-NAC breast MRI examination in our hospital from January 2016 to December2021 were retrospectively collected.On the basis of the pathological results of surgical specimens after neoadjuvant chemotherapy,all patients were classified as pCR group and non-PCR group.The areas of interest within and around the tumor were manually mapped by the radiologist using 3Dslicer software,and image omics features were extracted from ADC images and the Dynamic contrast-enhancement MRI(DCEMRI)phase I image for each patient,respectively.As there were few cases in the pCR group in this study,the samples were first balanced by the Synthetic minority oversampling technique(SMOTE),and then classified randomly as training group and testing group by the proportion of 7:3.After standardizing the data,In the training group,feature selection and dimension reduction were carried out through variance analysis,Spearman correlation analysis,Mann-Whitney test,Least absolute shrinkage and selection operator(LASSO).Then,logistic regression was used to establish models,including intratumoral,peritumoral and intratumoral combined peritumoral models with DCE and ADC sequences alone,and intratumoral,peritumoral and intratumoral combined peritumoral models with ADC+DCE sequences.The Receiver operating characteristic curve(ROC)can assess the predictive effectiveness of every model in the training group and test group for breast cancer pCR,and the best imaging omics model was selected and its Radscore was calculated.Then,univariate and multivariate Logistic regression analysis were used to determine the clinical features of all patients in the training group,and independent clinical predictors were screened out.The clinics-imaging model diagram for predicting breast cancer pCR was constructed by combining with the best imaging model Rad-score.ROC curve,calibration curve and decision curve(DCA)were used to asess and proof the effectiveness of the model.Results: The intratumoral and peritumoral models with DCE sequences alone produced similar predictive performance,with AUC values of 0.789 and 0.767 in the training group,respectively.The intratumoral model was superior to the peritumoral model,and the AUCs in the training group were 0.866 and 0.745,respectively.For ADC+DCE sequences,the intratumoral and peritumoral models also produced similar predictive performance,with AUCs of 0.892 and 0.843 in the training group,respectively.Based on all models,whether single or combined,compared with intratumoral and peritumoral models,intratumoral +peritumoral models always produced higher predictive efficiency,among which ADC+DCE intratumoral +peritumoral models had the best performance,with AUC values of 0.934 and 0.906 in the training group and test group,respectively.On the basis of univariate Logistic regression analysis of clinical characteristics,it was found that age,clinical lymph node status,ER,PR and HER2 were correlated with post-NAC pCR of breast cancer(P < 0.05),while only clinical lymph node metastasis status and HER2 were independent predictive factors of breast cancer by multivariate Logistic regression analysis(P < 0.05).Incorporate it into the construction of clinical models;HER2,lymph node status and the best imaging model Rad-score were combined to construct the clinical imaging model diagram.The AUC of the training group and the test group reached 0.963 and 0.955,indicating the best performance of the model.De Long’s test showed that in the training group,there was a statistically significant difference in AUC values among the clinical model,the clinical-imaging group diagram model and the imaging group model(P < 0.05,P values were 0.0002 and0.014 respectively).The calibration curve exhibit a good consistency among the prediction probability of pCR and the reality.The DCA shows that a net benefit can be obtained from the line graph model.Conclusion: Lymph node metastasis status and HER2 can be used as independent clinical predictors of post-NAC pCR of breast cancer.The clinics-imaging profile was constructed by combining them with the best imaging omics model Rad-score(ADC+DCE intra tumor combined with peritumoral),which has better forecast efficacy for pathological complete response of neoadjuvant chemotherapy in breast carcinoma.And can guide clinical treatment decision.