AI and Cancer Research ── Breast ── 2026-11-19
Breast cancer: treatment, trials and AI
- Abstract
Accurate risk stratification is essential for guiding treatment decisions and postoperative management in breast cancer. However, unimodal models are limited in their ability to fully exploit complementary information from clinical, radiological, and histopathological data associated with metastatic progression. We aimed to develop and validate Breast Cancer Bone Metastasis Network (BCBM-Net), a transformer-based multimodal deep learning network for predicting postoperative bone metastasis in patients with estrogen receptor (ER)-positive, human epidermal growth factor receptor 2 (HER2)-negative breast cancer. In this multi-institutional retrospective study, 1,240 patients from five clinical centers were included in training, internal test, and independent external test cohorts. BCBM-Net integrated selected clinicopathological variables, ultrasound (US)-derived features, and pathomics features extracted from whole-slide images (WSIs). Modality-specific fully connected encoders were used to learn nonlinear representations and project heterogeneous features into a shared latent space, followed by adaptive fusion using a multi-head attention transformer. Model performance was evaluated using receiver operating characteristic analysis, DeLong tests, and net reclassification improvement (NRI) analysis and was compared with clinical, US-based, and pathomics unimodal models. BCBM-Net achieved consistently higher discriminative performance than unimodal models across all cohorts, with area under the receiver operating characteristic curve (AUC) of 0.873, 0.866, and 0.855 in the training, internal test, and external test sets, respectively. DeLong tests and NRI analyses further supported the added predictive value of multimodal fusion. BCBM-Net effectively integrates routinely available clinical, US, and histopathological information to provide individualized assessment of 5-year postoperative bone metastasis risk. Its consistent performance in independent external centers supports the value of transformer-based multimodal fusion for cancer prognostication. BCBM-Net could provide a practical framework for risk-adapted surveillance and postoperative management, although prospective and multinational validation is required before clinical implementation.
Journal IF-equivalent: 7.9 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-05; not the official Clarivate IF)Reference: Zhang Z, Huang G, Deng Z, Ouyang L, Wu Z, Zhang K, et al. Transformer-based multimodal deep learning network for predicting postoperative bone metastasis in ER-positive/HER2-negative breast cancer: a multi-institutional retrospective study. BMC Med. 2026 Nov 16 [Epub ahead of print]. doi:10.1186/s12916-026-05352-3.Checked: Abstract only - Abstract
HER2-low breast cancer has emerged as a distinct subtype with unique biological characteristics. Habitat imaging can characterize intratumoral heterogeneity by partitioning tumors into subregions with distinct imaging phenotypes. However, the comparative prognostic value of habitats derived from diffusion-weighted imaging (DWI) and dynamic contrast-enhanced MRI (DCE-MRI) remains unexplored. This study aimed to develop and compare DWI and DCE habitat radiomics models for predicting disease-free survival (DFS) in HER2-low breast cancer. A total of 401 patients from four centers were retrospectively included. DWI and DCE images were independently clustered into three habitats using K-means, relabeled by signal intensity as low (L), medium (M), and high (H). From each habitat, 107 three-dimensional radiomics features were extracted. LASSO-Cox regression with 5-fold cross-validation was used for model construction. Model performance was assessed by C-index, time-dependent AUC, Brier score, and log-rank test. Among pure imaging models, the DCE high-signal habitat (DCE_H) demonstrated the best performance (C-index = 0.642, P = 0.033). When combined with clinical variables, the DCE_H model achieved modestly higher time-dependent AUC on nested cross-validation (3-year AUC 0.699→0.720; 5-year AUC 0.728→0.754), although this incremental difference did not reach statistical significance in paired comparison ( P = 0.056 and 0.080, respectively). Standard multivariable Cox regression suggested that DCE_H Sphericity was associated with DFS (hazard ratio [HR] = 0.067, 95% confidence interval [CI]: 0.010–0.520, P = 0.0096). In contrast, DWI habitat features were not selected by LASSO when clinical variables were present. DCE high-signal habitat radiomics—particularly Sphericity—was associated with DFS in HER2-low breast cancer, although its incremental value over clinical variables was modest and not statistically significant. DWI features added little. External validation is required. The study was retrospectively registered in the Chinese Clinical Trial Registry (ChiCTR2400085188, registered June 03, 2024, http//www.chictr.org.cn/).
Journal IF-equivalent: 4.8 (OpenAlex 2-year mean citedness, value as of 2026-11-12, retrieved 2026-11-13; not the official Clarivate IF)Reference: Meng M, Liu C, Peng Y, Wang F, Huang R, Zhu M, et al. MRI habitat radiomics with clinical features for prognosis prediction in HER2-low breast cancer: a multicenter study. BMC Cancer. 2026 Nov 16 [Epub ahead of print]. doi:10.1186/s12885-026-17380-6.Checked: Abstract only