Evaluation of Multiparametric MRI Radiomics-Based Nomogram in Prediction of Response to Neoadjuvant Chemotherapy in Breast Cancer: A Two-Center study

This study established and validated two radiomics-based nomograms for predicting significant remission and pCR to neoadjuvant chemotherapy in breast cancer patients. DCE-MRI and clinical data of 387 patients were retrospectively collected to bulid two nomograms, which performed well with AUC values of 0.86 and 0.80 in external validation cohorts. DCA showed the nomograms obtained the most clinical benefit. Introduction:This study evaluated the performance of primary foci of breast cancer on multiparametric magnetic resonance imaging (MRI) contributing to establish and validate radiomics-based nomograms for predicting the different pathological outcome of breast cancer patients after neoadjuvant chemotherapy (NAC). Materials and Methods:Retrospectively collected 387 patients with locally advanced breast cancer, all treated with NAC and received breast dynamic contrast-enhanced MRI (DCE-MRI) before NAC. Radiomics signatures were extracted from region of interest (ROI) on multiparametric MRI to build rad score. Clinical-pathologic data and radiological features established the clinical model. The comprehensive model featured rad-score, predictive clinical-pathologic data and radiological features, which was ultimately displayed as a nomogram. Patients were grouped in 2 different ways in accordance with the Miller-Payne (MP) grading of surgical specimens. The first grouping method: 181 patients with pathological reaction grades IV -V were included in the significant remission group, while 206 patients with pathological reaction grades I -III were included in the nonsignificant remission group. The second grouping method: 117 patients with pathological complete response (pCR) were assigned to the pCR group, and 270 patients who failed to meet pCR were assigned to in the non-pCR group. Two combined nomograms are created from 2 grouped data for predicting different pathological responses to NAC. The area under the curves (AUC) of the receiver operating characteristic curves (ROC) were used to evaluate the performance of each model. While decision curve analysis (DCA) and calibration curves were used for estimating the clinical application value of the nomogram. Results:Two combined nomograms embodying rad score and clinical-pathologic data outperformed, showing good calibrations for predicting response to NAC. The combined nomogram predicting pCR showed the best performance with the AUC values of 0.97, 0.90 and 0.86 in the training, testing, and external validation cohorts respectively. The AUC values of another combined nomogram predicting significant remission: 0.98, 0.88 0.80 in the training, testing and external validation cohorts. DCA showed the comprehensive model nomogram obtained the most clinical benefit. Conclusions:The combined nomogram could preoperatively predict significant remission or even pCR to NAC in breast cancer based on multiparametric MRI and clinical-pathologic data.