ABSTRACT
Objective:To investigate the value of delta radiomics based on longitudinal changes of dynamic contrast enhanced MRI (DCE-MRI) in predicting pathological complete response (pCR) after neoadjuvant therapy (NAT) for breast cancer.Methods:The clinicopathological and imaging data of 117 patients with breast cancer confirmed by surgical pathology from April 2019 to November 2021 at Jiangxi Cancer Hospital were analyzed retrospectively. All patients were female with 23?74 (48±10) years old. The patients were randomly divided into training (81 cases) and test sets (36 cases) at the ratio of 7∶3 according to the number of random seeds in the software. All patients underwent DCE-MRI before and after early NAT (2 courses). The maximum diameter relative regression value of breast tumors before and after early NAT (D%) was calculated and used to construct a conventional imaging model. The delta radiomic features were extracted based on pre-NAT and early-NAT (2 courses) DCE-MRI and selected by redundancy analysis and least absolute shrinkage and selection operator algorithm. A ten-fold cross-validation method was used to construct the delta radiomic model and Radscore was calculated for each patient. All patients were classified into pCR group and non-pCR group according to the surgical pathology after NAT. Significant clinicopathological variables were selected by univariate analysis and stepwise regression method. They were integrated with D% and Radscore to build the combined model and nomogram. The model performance in predicting pCR after NAT in breast cancer was evaluated by the receiver operating characteristic curve and the area under the curve (AUC), and the clinical utility of the models was compared by using clinical decision curves.Results:The combined model had the best diagnostic performance among the three models, with an AUC of 0.90 in the training set and 0.87 in the test set. The Radscore had the highest weight in the nomogram. In the training set, the diagnostic performance of the combined model and delta radiomics model were better than that of the conventional imaging model ( Z=?3.48, P=0.001; Z=2.54, P=0.011). The clinical decision curves showed an overall greater clinical benefit of the combined model compared with the conventional imaging model and delta radiomic model. Conclusions:The addition of significant clinicopathological variables and Radscore of delta radiomic model which represents the longitudinal changes in tumor heterogeneity to the conventional imaging model may improve the predictive ability of pCR. The delta radiomic may serve as a noninvasive biomarker for early prediction of NAT response.