Abstract P2-06-18: Deep learning-based tumor microenvironment signature predicts progression-free survival in early-stage hormone receptor positive and HER2 negative breast cancer

Firas Khader, Omar S. M. El Nahhas, Evan D. Paul, Pavol Čekan, Nicoleta Zenovia Antone, Daniel Truhn, Jakob Nikolas KatherClinical 2025 · DOI

Abstract

Abstract Molecular risk assays, such as OncotypeDX, have emerged as a robust method for guiding adjuvant chemotherapy alongside endocrine therapy in early-stage hormone receptor positive (HR+) and HER2 negative (HER2-) breast cancer (BC). However, these assays have limitations in that they lack the capability to capture the heterogeneity and complexity of the tumor microenvironment. They do not account for morphological features in routine hematoxylin and eosin (H&E)-stained histopathology slides—features that can provide insights into how the tumor cells are behaving in situ. Additionally, molecular assays are expensive and time-consuming, limiting their widespread use in clinical practice. Consequently, there is a need for alternative or complementary prognostic tools that can analyze tumor morphology and the TME from H&E slides. Deep learning (DL) can predict cancer biomarkers directly from H&E slides, enabling the automated and consistent quantification of the tumor microenvironment (TME) for clinical decision-making. Our study focuses on the prediction of TME signatures on a large international and multi-centric cohort for the prognostication of BC patients. We developed a weakly-supervised, transformer-based multi-task deep regression model to predict signatures of the TME directly from routine H&E-stained histology slides. We trained the model using 5-fold cross-validation on a large pan-cancer cohort of patients (n=2,690), and validated its prognostication performance on an external cohort of early-stage HR+/HER2- female BC patients (n=470). The model predicted signatures of tumor cell proliferation and tumor infiltrating lymphocytes (TILs) as a continuous variable, which was carried forward for survival analyses as a composite interaction score. Using longitudinal data from the multi-centric external cohort (median follow-up = 133 months), survival analysis was performed using a Cox proportional hazards model with adjustment for age, nodal status, stage, administration of adjuvant chemotherapy, and a research-based molecular risk of recurrence score. Our AI-based TME signature was significantly associated with improved 12-year progression-free survival, with a hazard ratio = 2.28, (95% confidence interval; 1.19, 4.36), p=0.01. Our new AI-based prognostication method, which we call SAIBR (StratifAI BReast), provides a cheap and readily available algorithm for analyzing digital pathology images of H&E sections of BC to quantify tumor cell proliferation and TILs. The data show that SAIBR is an informative prognostic marker for 12-year progression-free survival and that it provides prognostic information beyond standard clinicopathological factors and validated molecular risk assessment in early-stage HR+/HER2- breast cancer. Citation Format: Firas Khader, Omar S. M. El Nahhas, Firas Khader, Evan Paul, Pavol Cekan, Nicoleta Antone, Daniel Truhn, Jakob Nikolas Kather. Deep learning-based tumor microenvironment signature predicts progression-free survival in early-stage hormone receptor positive and HER2 negative breast cancer [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P2-06-18.
BibTeX
@inproceedings{khader2025abstract,
  author = {Firas Khader and Omar S. M. El Nahhas and Firas Khader and Evan D. Paul and Pavol Čekan and Nicoleta Zenovia Antone and Daniel Truhn and Jakob Nikolas Kather},
  title = {{Abstract P2-06-18: Deep learning-based tumor microenvironment signature predicts progression-free survival in early-stage hormone receptor positive and HER2 negative breast cancer}},
  booktitle = {Clinical},
  year = {2025},
  doi = {10.1158/1557-3265.sabcs24-p2-06-18},
}

Loading related content...

Figure