Radiomics Based On Pharmacokinetic Parametric Maps From DISCO DCE-MRI Predicts Pathologic Response In Breast Cancer Patients Treated With Neoadjuvant Chemotherapy

Objective:Differential Subsampling with Cartesian Ordering（DISCO）is a dynamic contrast-enhanced MRI（DCE-MRI）imaging technique with high spatial and temporal resolution.The aim of this study is to evaluate the value of the radiomics features extracted by the pharmacokinetic parameter maps from DISCO DCE-MRI and delta-radiomics for predicting the response to neoadjuvant chemotherapy（NAC）.Methods:142 breast cancer neoadjuvant chemotherapy patients who received breast DISCO DCE-MRI scan before NAC（pre-NAC）and after the first cycle of NAC（1st-NAC）were prospectively included from May 2019 to May 2022.Patients were divided7:3 into training and validation cohorts based on their admission time.The pharmacokinetic parameters maps of Volume transfer constant(K^trans),Volume fraction of plasma（V_p）,and Rate constant(K^ep)were automatically computed using the reference region model in Omni-Kinetic software.The radiomics features based on the full tumor volume were extracted from the DCE-MRI quantitative parameter map,and the delta-radiomics features,defined as changes from pre-NAC to 1st-NAC,were calculated.Pre-,1st-and delta-radiomics models and longitudinal fusion models were constructed based on logistic regression.Finally,a clinical-radiomics model was built by combining radiomic features and important clinical factors.The effectiveness of each model was evaluated by plotting the receiver operating characteristiccurve（ROC）and calculating the area under the ROC curve（AUC）,and the predictive performance was compared among the models using Delong test.Results:Among the three individual radiation models,the delta-radiomics model showed significant performance in the training（AUC;0.802,95%CI:0.710,-0.876）and validation set（AUC:0.785,95%CI:0.633,0.895）.The predictive performance of the longitudinal fusion model was slightly improved compared to the individual models.After adding clinicopathological factors,all clinical-radiomics models had AUC values above 0.8 in the training set,and in the validation set,the highest AUC value of the clinical-radiomics model was 0.840.Clinical-radiomics models were helpful to distinguish p CR from non-p CR.Conclusion:Radiomic based on pharmacokinetic parametric maps from DISCO DCE-MRI could be used as a predictor for NAC response,and the combination of radiomics with clinicopathology information improves the predictive efficacy of NAC.