MRI-based Radiomics For Prediction Of Tumor Regression Pattern To Neoadjuvant Chemotherapy In Breast Cancer

BackgroundNeoadjuvant chemotherapy(NAC)has become more commonly used in patients with locally advanced breast cancer.The primary benefit of NAC is to preoperatively downstage the breast tumor and increase the likelihood of breast-conserving surgery(BCS).The regression pattern of tumor after NAC would affect the choice of surgical approach.Therefore,pretreatment prediction of tumor regression pattern can assist clinician to optimize individualized therapeutic strategy for better outcome.Objectives1)To investigate the predictive value of the combination of pretreatment MRI qualitative features and clinicopathologic features for early prediction of tumor regression pattern to NAC in breast cancer.2)To further investigate whether quantitative MRI-based radiomics features combined with qualitative features and clinicopathological features could improve the accuracy of tumor regression pattern prediction.Methods1)We retrospectively analyzed 420 patients who received NAC and underwent definitive surgery in our hospital from February 2012 to August 2020.Pathologic findings of surgical specimens were used as the gold standard to classify tumor regression patterns into concentric and non-concentric shrinkage.Morphologic and kinetic MRI features were both analyzed.Univariable and multivariable analyses were performed to select the key clinicopathologic and MRI features for pretreatment prediction of regression pattern.Logistic regression and six machine learning methods were used to construct prediction models,and their performance were evaluated with receiver operating characteristic(ROC)curve.2)Study population were the same as the Part 1 with a total of 420 cases and were divided into a training cohort(n=294)and validation cohorts(n=126)in the ratio of 7:3 according to the chronological order of MRI examination.Based on dynamic contrast-enhanced MRI,radiomics features were extracted.After feature selection,we used artificial neural networks to construct a radiomics signature.Then,combined with MRI qualitative features and clinicopathologic features to construct the radiomic combined model,and its prediction performance was evaluated with ROC curve,calibration curve and decision curve analysis.Results1)Two clinicopathologic variables and three MRI features were selected as independent predictors to construct prediction models.The apparent area under curves(AUCs)of seven prediction models were in the range of 0.669-0.740.The logistic regression model yielded an AUC of 0.708(95%CI:0.658-0.759),and the decision tree model achieved the highest AUC of 0.740(95%CI:0.6910,787).For internal validation,the optimism-corrected AUCs of seven models were in the range of 0.592-0.684.There was no significant difference between the AUCs of the logistic regression model and that of each machine learning model.2)The MRI-based radiomics signatures including eight features.In the training cohort and validation cohort,the radiomics signature model yielded an AUC of 0.738(95%CI:0.705-0.754)and 0.696(95%CI:0.585-0.712),respectively;the clinical predictive model yielded an AUC of 0.676(95%CI:0.636-0.741)and 0.619(95%CI:0.601-0.716),respectively;the combined predictive model yielded an AUC of 0.802(95%CI:0.753-0.824)and 0.764(95%CI:0.6850.820),respectively.Decision curve analysis showed the clinical use of the combined predictive models.To facilitate clinical applications,we further visualized the combined predictive model as a nomogram.ConclusionsPrediction models combining pretreatment quantitative MRI-based radiomics features and qualitative MRI features and clinicopathologic features are useful for predicting tumor regression pattern in breast cancer,which can assist to select patients who can benefit from NAC for de-escalation of breast surgery and optimize individualized therapeutic strategy for better outcome.