Integrating MRI radiomics and deep learning for predicting pathologic complete response to neoadjuvant chemotherapy in breast cancer: a multicenter study.
Authors
Affiliations (7)
Affiliations (7)
- Cancer Center, Department of Ultrasound Medicine, Zhejiang Provincial People's Hospital, Affiliated People's Hospital, Hangzhou Medical College, Hangzhou, Zhejiang, China.
- Cancer Center, Department of Radiology, Zhejiang Provincial People's Hospital, Affiliated People's Hospital, Hangzhou Medical College, Hangzhou, Zhejiang, China.
- Department of Ultrasound，Sichuan Provincial People's Hospital, Chengdu, China.
- Department of Radiology, Guangdong Provincial People's Hospital (Guangdong Academy of Medical Sciences), Southern Medical University, Guangzhou, 510080, China.
- Advanced Analytics, GE Healthcare, Hangzhou, China.
- Department of Medical Ultrasonics, The First Affiliated Hospital of Guangzhou Medical University, 151 Yanjiang West Road, Guangzhou, 510120, China.
- Cancer Center, Department of Radiology, Zhejiang Provincial People's Hospital, Affiliated People's Hospital, Hangzhou Medical College, Hangzhou, Zhejiang, China. [email protected].
Abstract
Develop a multimodal fusion model combining MRI radiomics and deep learning (DL) to predict pathologic complete response (pCR) in breast cancer patients post-neoadjuvant chemotherapy (NACT). Patients with locally advanced breast cancer from two centers (Center 1: 404 training, Center 2: 174 testing) were enrolled. Pretreatment MRI radiomic features and DL-derived features were extracted, selected via Minimum Redundancy Maximum Relevance and Least Absolute Shrinkage and Selection Operator, and integrated with clinical risk factors into a joint multimodal fusion model using XGBoost. The model was validated internally on an independent test cohort and externally on the Duke Breast Imaging Dataset (284 patients) to assess generalizability across diverse populations and imaging protocols. The fusion model achieved AUCs of 0.935 (training), 0.916 (testing), and 0.83 (external validation), surpassing standalone radiomics and DL models. External validation confirmed robustness to institutional variability, with strong calibration and clinical utility. The XGBoost-integrated radiomics-DL-clinical model accurately predicts pCR, validated across independent cohorts, and may optimize personalized treatment strategies.