AI-based quantification of TILs using hematoxylin and eosin and immunohistochemistry-stained slides in triple-negative breast cancer.
Takuma Kobayashi,Nobumoto Tomioka,Kanako C Hatanaka,Teppei Konishi,Mateusz Grynkiewicz,Daisuke Komura,Shumpei Ishikawa,Kenichi Watanabe,Masato Takahashi,Yutaka Hatanaka
DOI: https://doi.org/10.1200/jco.2024.42.16_suppl.e13608
IF: 45.3
2024-05-31
Journal of Clinical Oncology
Abstract:e13608 Background: The importance of accurately understanding the status of tumor-infiltrating lymphocytes (TILs) in the preoperative chemotherapy context, for triple-negative breast cancer is underscored, highlighting the dual role of TILs in gauging treatment efficacy and prognostication. However, the difficulty in quantitatively and objectively assessing TILs in clinical practice is acknowledged, noting the barrier to its adoption as a standard diagnostic end-point. This study proposes a novel AI-based method to overcome these challenges, by quantitatively assessing TILs using hematoxylin and eosin (H&E) and immunohistochemistry (IHC) staining, and exploring their prognostic relationship. Methods: In this study, 68 cases of breast cancer with axillary lymph node metastasis at the time of resection, were selected from a total of 3,902 resections conducted between 2002 and 2016. To quantitatively assess TILs, we developed a quantitative assessment method, integrating an AI-based epithelial detection model and non-rigid image registration techniques to align H&E and IHC-stained images for precise measurement of immune markers (CD3, CD8, FOXP3, PD-L1) within the tumor stroma. Using an AI-based model trained on 12,273 annotated epithelial images, the stroma, excluding the epithelium, was identified. The tumor bed was also defined by a pathologist. Finally, we assessed the prognostic significance of the TILs for disease-free survival (DFS) through ROC analysis and log-rank tests. Results: Optimal thresholds for CD3 (0.152 mm2), CD8 (4.8 mm2), FOXP3 (0.25 mm2), and PD-L1 (0.115 mm2) were identified, with corresponding AUCs of 0.678, 0.625, 0.746, and 0.659. Patients with cell counts or areas above these thresholds demonstrated improved DFS (P<0.05 for all markers). Conclusions: These findings suggest that the quantification of immune cells within the tumor bed provides reliable prognostic indicators for recurrence. The AI-based TILs quantification offers a significant advancement in the prognostic assessment of triple-negative breast cancer. In addition, it improves the objectivity and clinical utility of pathological evaluations. This precise measurement of TILs could be transformative in patient prognostication and the formulation of treatment plans.
oncology