Tumor Habitat and Global Radiomics on DCE-MRI-Derived Metrics Enable Visualization and Quantification of Breast Cancer Spatio-temporal Hemodynamic Heterogeneity

Authors:
  • YRYa Ren
    Ya Ren
  • XTXuetong Tao
    Xuetong Tao
  • LLLin Li
    Lin Li
  • MWMeng Wang
    Meng Wang
  • SCShuluan Chen
    Shuluan Chen
  • JWJie Wen
    Jie Wen
  • SFSha Feng
    Sha Feng
  • WCWei Cui
    Wei Cui
  • YGYuan Guo
    Yuan Guo
  • DLDong Liang
    Dong Liang
  • HZHairong Zheng
    Hairong Zheng

Abstract
Background Breast tumors exhibit substantial intratumoral hemodynamic heterogeneity associated with diagnostic uncertainty, yet conventional radiomics relies on whole-lesion analysis and does not capture spatially distinct functional subregions. This study aimed to develop a quantitative approach to visualize and quantify intratumor spatio-temporal hemodynamic heterogeneity and validate its clinical utility by distinguishing benign and malignant breast lesions. Methods 805 patients with preoperative breast MRI scans were retrospectively collected from two academic institutions between June 2018 and December 2023. Tumor voxels within manually segmented lesions were stratified by wash-in rate and wash-out enhancement, and K-means clustering enabled subregional habitat segmentation. Radiomics features were extracted from both the whole tumor and each habitat. Seven predictive models were developed using conventional radiomics features, habitat radiomics features, or their combinations derived from wash-in and wash-out maps. Model performance was evaluated using the area under the receiver operating characteristic curve, and comparisons were performed using the Friedman test. This framework integrates dynamic contrast–based habitat analysis with semi-quantitative parametric map-based radiomics, aiming to better capture intratumoral hemodynamic heterogeneity. Findings The study included 805 women (49 years ± 12) with 923 breast lesions (652 malignant and 271 benign). A total of 594 patients were assigned to the training and internal validation cohorts, and 211 patients were assigned to the external validation cohort. The combined model achieved the highest diagnostic performance, with AUCs of 0.920, 0.896, and 0.952 in the training, internal validation, and external validation cohorts, respectively, which significantly outperformed both the wash in-out radiomics model (AUCs: 0.899, 0.877, 0.911; p<0.001 ) and the wash in-out habitat model (AUCs: 0.861, 0.836, 0.923; p<0.001). Interpretation Tumor habitat and global radiomics based on DCE-MRI–derived kinetic metrics enable visualization and quantification of intratumoral hemodynamic heterogeneity and improve diagnostic performance for differentiating benign and malignant breast lesions.