Prediction of pathological complete response to neoadjuvant therapy in breast cancer integrating intratumoral and peritumoral delta radiomics with clinical features: a study based on multiparametric MRI.
Authors
Affiliations (5)
Affiliations (5)
- Department of Radiology, The Affiliated Hospital of Qingdao University, No.59 Haier Road, Qingdao, 266000, China.
- Medical College of Qingdao University, Qingdao, China.
- Department of Thyroid Surgery, The Affiliated Hospital of Qingdao University, Qingdao, China.
- Department of Radiology, The Affiliated Hospital of Qingdao University, No.59 Haier Road, Qingdao, 266000, China. [email protected].
- Department of Breast Imaging, The Affiliated Hospital of Qingdao University, 59 Haier Road, Qingdao, China. [email protected].
Abstract
To develop a combined model integrating intratumoral and peritumoral delta-radiomics from multi-parametric MRI with clinical features for predicting pathological complete response (pCR, i.e., Miller-Payne grade V response) to neoadjuvant chemotherapy (NAT) in breast cancer. A total of 254 patients with breast cancer from two hospitals were retrospectively included. Radiomics features were extracted from both intratumoral and peritumoral regions on multi-parametric MRI obtained at baseline and after the second cycle of NAT, and delta radiomics features were subsequently derived to quantify longitudinal changes between the two time points. Six machine learning algorithms were employed to construct and compare delta-radiomics and clinical models to identify the optimal algorithm. The integrated model combining intratumoral and peritumoral delta radiomics and clinical features was developed based on the optimal algorithm to predict pCR. Additionally, the Shapley Additive Explanations (SHAP) method was applied to interpret the contributions of features within the optimal model. Compared with other machine learning algorithms, ExtraTrees algorithm achieved superior predictive performance for both delta radiomics and clinical models. Receiver operating characteristic and decision curve analysis demonstrated that the multimodal model outperformed the delta radiomics and clinical models in predicting pCR. Furthermore, SHAP analysis revealed that dynamic changes in intratumoral and peritumoral radiomics features during neoadjuvant therapy jointly drive the predictive performance of the fusion model. This study highlighted the potential of intratumoral and peritumoral regions in delta radiomics analysis to evaluate neoadjuvant therapy for breast cancer, thereby further broadening the prospects for personalized decision-making in breast cancer treatment.