Multiparametric MRI-based Radiocics Analysis For Prediction Of Breast Cancers Insensitive To Neoadjuvant Chemotherapy

Purpose:Neoadjuvant chemotherapy(NAC)has become an important part of standard breast cancer treatment and is currently widely used in early or locally advanced breast cancer.However,breast cancer is a highly biologically heterogeneous tumor,and the response to neoadjuvant chemotherapy varied widely.However,a few breast cancers were found not to be sensitive to NAC and even had larger tumors after NAC.Therefore,there is a need to find a effective approach to accurately separate the non-responders from the responders in the early stage.This study intends to evaluate the value of multiparametric magnetic resonance imaging(MRI)in pretreatment prediction of breast cancers insensitive to neoadjuvant chemotherapy(NAC)and construct a predictive model that integrates clinical,pathological,and imaging information to explore imaging phenomenological features and NAC efficacy.Materials and Methods:This study retrospectively enrolled 125 patients with breast cancer who underwent multiparametric magnetic resonance imaging(MRI)before NAC.The patients were assigned a 1:1 ratio according to the date of diagnosis of breast cancer to the training cohort and the validation cohort,with 63 patients in the training cohort and 62 patients in the validation cohort.All patients underwent surgery after completion of neoadjuvant chemotherapy.The pathological Miller-Payne grading system was used to evaluate the surgically resected tumor to assess the response of breast cancer to NAC.Grade 1-2 cases were classified as insensitive to NAC.We extracted 1941 features from preoperative MRI imaging of neoadjuvant chemotherapy in training cohort,including dynamic contrast-enhanced(DCE),T2-weighted imaging(T2WI)and diffusion-weighted imaging(DWI)MRI sequences.After feature selection,the optimal feature set was used to construct a radiomic signature using machine learning.Then,the multivariate logistic regression analysis was used to select the clinical factors with independent predictive power.Firstly,a clinical predictive model was established by independent clinical risk factors.Then a combined predictive model was established by incorporating the radiomic signature and independent clinical risk factors.The performance of the combined model was assessed with the results of independent validation.Results:Four features were selected for the construction of the radiomic signature based on the primary cohort.Then,the independent clinical risk factors selected by multivariable logistic regression to establish a clinical prediction model.Finally,we combined the radiomics with the HER2 state and Ki67 expression state to establish the combined prediction model,the combined prediction model shows better predictive power than the clinical prediction model.In the validation cohort,the area under the receiver-operating characteristic curve(AUC)of the combined prediction model is 0.935(95%confidence interval,0.848-1),the accuracy was 93.55%(91.44%-95.69%),the positive predictive value was 75.00%(60.51%-90.29%)?and the negative predictive value was 94.83%(92.81%-96.85%).and its clinical utility was confirmed by the decision curve analysis.Conclusion:The radiomic signature established in our study could be used to predict the response of breast cancer to NAC before treatment.The combined predictive model combined with independent clinical risk factors further improved the predictive power.It has potential in predicting drug insensitive breast cancers.