AI and Cancer Research ── Breast ── 2026-10-30
Breast cancer: treatment, trials and AI
- Abstract
To develop and compare machine-learning (ML) models based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) radiomics, clinical variables, and their combination for predicting pathological complete response (pCR) after neoadjuvant chemotherapy (NAC) in breast cancer, and to critically evaluate the stability of each model in an independent test cohort. This retrospective single-center study included 272 patients treated between May 2020 and May 2025 who were retrospectively analyzed. pCR was determined using the Miller-Payne grading system, corresponding to ypT0/Tis ypN0 (no residual tumor in sampled axillary nodes). In total, 1,834 features were extracted from pretreatment phase-3 DCE-MRI with PyRadiomics. All feature selection (Mann–Whitney U test, Spearman redundancy pruning, LASSO) and hyperparameter tuning were performed exclusively in the training cohort ( n = 217); the held-out test cohort ( n = 55) was used once for final evaluation. LR, a radial-basis-function (RBF) SVM, and an MLP were compared, together with a clinical signature (HER2 and axillary nodal status) and a radiomics–clinical nomogram; discrimination (AUC) and decision curve analysis (DCA) were assessed. Of 272 patients, 98 (36.03%) achieved pCR. Mann–Whitney screening and Spearman pruning retained 34 features, from which LASSO selected 17 with non-zero coefficients. The SVM-based radiomics model performed best (training AUC: 0.873; test AUC: 0.782). The clinical signature was the most stable model (training AUC 0.885; test AUC 0.806), whereas the nomogram, despite the highest training AUC (0.901), showed a marked decline in the test cohort (AUC 0.686, 95% CI 0.350–1.000), and its DCA advantage was not reproduced in the small test cohort. HER2 positivity and node-negative disease robustly identified patients likely to achieve pCR, and MRI radiomics provided moderate complementary discrimination; however, the incremental benefit of the combined nomogram was not confirmed in the independent test cohort, indicating possible overfitting. The radiomics-based models require stratified or external validation before clinical deployment.
Journal IF-equivalent: 3.8 (OpenAlex 2-year mean citedness, value as of 2026-10-02, retrieved 2026-10-03; not the official Clarivate IF)Reference: Li H, Zhang L, Zeng Y, Ye Z, Zhang Y. MRI-based radiomics analysis: comparative application of different machine learning models in predicting pathological complete response after neoadjuvant chemotherapy for breast cancer. BMC Med Imaging. 2026 Oct 28 [Epub ahead of print]. doi:10.1186/s12880-026-02917-9.Checked: Abstract only - Abstract
Background: Machine learning-based assessment of tumor-infiltrating lymphocytes (TILs) has been proposed to improve the reproducibility and clinical utility of immune-contexture evaluation in breast cancer histopathology. However, existing studies differ substantially in task definition, annotation strategy, model architecture, validation design, and clinical endpoint. This systematic review and evidence-gap analysis aimed to map the translational readiness of TIL-related machine learning studies in breast cancer histopathology and to identify cluster-specific methodological gaps.
Methods: A systematic search of PubMed, Web of Science, Scopus, ScienceDirect, and Google Scholar was conducted up to 20 April 2026, supplemented by backward citation tracing and a targeted Google search of the 30 most relevant additional results. Eligible studies included breast cancer histopathology-based machine learning, deep learning, artificial intelligence, or computational image-analysis studies evaluating TILs or TIL-related immune-cell features. Studies were grouped by principal application into detection/classification, outcome prognostication, and treatment-response prediction. Extracted variables included data source, specimen type, model family, annotation strategy, validation design, clinical comparator use, calibration, decision-curve analysis, incremental value, reproducibility, and methodological challenges. Because of heterogeneity in endpoints and performance metrics, synthesis was descriptive and evidence-map based.
Results: The primary evidence map included 51 histopathology-image studies: 19 detection/classification, 18 outcome prognostication, and 14 treatment-response prediction studies. External validation was reported in 31.6% of detection/classification studies, 61.1% of prognostication studies, and 57.1% of treatment-response prediction studies. Clinical comparator reporting increased across the same continuum, from 63.2% to 94.4% and 92.9%, respectively. However, calibration was reported in only one prognostication study (5.6%) and in no detection/classification or treatment-response prediction studies. Decision-curve analysis was similarly limited to one prognostication study. Class imbalance was near-universal (94.7–100% across clusters). Scanner/stain robustness was addressed in a majority of studies (63.2–83.3%), but fairness or subgroup analyses remained inconsistently reported (27.8–57.1%).
Conclusions: TIL machine learning in breast cancer is technically active but clinically under-validated. Future studies should move beyond model development toward externally validated, calibrated, clinically benchmarked, reproducible, and decision-relevant evidence, using cluster-specific planning standards for detection, prognostication, and treatment-response prediction. Systematic review registration https://www.crd.york.ac.uk/PROSPERO/view/CRD42024574566 , identifier CRD42024574566.
Journal IF-equivalent: 2.7 (OpenAlex 2-year mean citedness, value as of 2026-10-02, retrieved 2026-10-03; not the official Clarivate IF)Reference: Dunenova G, Sarsembayev A, Oshibayeva A, Kyrykbayeva S, Abdrakhmanova Z, Omarova B, et al. Machine learning-based assessment of tumor-infiltrating lymphocytes in breast cancer histopathology: a systematic review and evidence-gap analysis of translational readiness. Front Oncol. 2026 Oct 26;16:1908349. doi:10.3389/fonc.2026.1908349.Checked: Abstract only - Abstract
Oncotype DX® (ODX) guides adjuvant chemotherapy decisions in early breast cancer (eBC), but its high-cost limits access in resource-constrained settings. We developed and evaluated an interpretable machine learning model using routinely available clinicopathologic variables to predict ODX recurrence score (RS) and optimize genomic testing. We analyzed 935 Brazilian patients with eBC from the GBECAM cohort (2005–2024). Ridge regression was selected for its predictive performance, interpretability, and computational efficiency. Patients were stratified into risk groups based on predicted probability of RS > 25, and recurrence-free survival was estimated using the Kaplan–Meier method. The model showed good discrimination for identifying RS > 25 (AUROC 0.818 [95% CI, 0.780–0.852]), with sensitivity of 0.556 and specificity of 0.859. The highest-risk quartile contained 60.8% of all RS > 25 cases (observed probability 42.8%), whereas the lowest-risk half had an observed probability of only 4.4%. These findings suggest that up to 75% of patients could be triaged without confirmatory ODX testing, reserving genomic assays for an intermediate-risk subgroup. The Ridge model outperformed previously published clinicopathologic models and may represent a cost-effective strategy to improve ODX testing prioritization and resource allocation.
Journal IF-equivalent: 6.2 (OpenAlex 2-year mean citedness, value as of 2026-10-08, retrieved 2026-10-09; not the official Clarivate IF)Reference: de Araújo JAP, de Carvalho Oliveira LJ, Los B, Araújo DC, Souza PH, Rosa DD, et al. Interpretable machine learning models for oncotypedDX® prediction using clinicopathologic data: a brazilian cohort study. npj Breast Cancer. 2026 Oct 26 [Epub ahead of print]. doi:10.1038/s41523-026-01065-1.Checked: Abstract only