MRI subregional analysis of spatial heterogeneity across residual cancer burden levels in breast cancer

作者：Yi Qin, Tingfeng Zhang, Chuqiao Luo, Jinhua Wang, Lin Li, Hong Hu, Jie Ma · 发表于：Applied Radiation and Isotopes · 年份：2025 · DOI：10.1016/j.apradiso.2025.112359 · 被引用次数：1 · 研究领域：Radiomics and Machine Learning in Medical Imaging、Ferroptosis and cancer prognosis、MRI in cancer diagnosis

BACKGROUND: Neoadjuvant chemotherapy (NAC) downstage breast cancer and increases breast-conserving surgery rates, but response variability challenges early efficacy prediction. While Residual Cancer Burden (RCB) is a robust post-NAC prognostic marker, its imaging and molecular determinants remain unclear. We investigated MRI-based tumor spatial heterogeneity related to RCB classification and identified radiogenomic biomarkers of treatment response. METHODS: Retrospective analysis of 375 breast cancer patients receiving NAC and surgery, stratified into RCB-0/I and RCB-II/III groups from two centers. Pre-treatment dynamic contrast-enhanced MRI enabled radiomic feature extraction and subregional analysis via unsupervised clustering. A support vector machine classifier predicted RCB classes. Radiogenomic associations integrated transcriptomic data using weighted gene co-expression network analysis (WGCNA), gene perturbation similarity analysis and Comparative Toxicogenomics Database. Meta-analysis of 21 cohorts evaluated GATA3 prognostic value. RESULTS: The radiomics model yield AUCs of 0.92, 0.87 and 0.87 in training set, test set and external validation set, respectively. High-RCB tumors exhibited greater texture heterogeneity, irregular shape, and signal variance. Radiogenomic correlations identified GATA3, CXCR4, CCND1, and VEGF as key genes linked to imaging phenotypes. GATA3 downregulation correlated with aggressive radiomic features and high RCB, confirmed by GPSA. Compara...