Deep learning-based quantitative assessment of tumor-infiltrating lymphocytes from hematoxylin and eosin-stained slides in triple-negative breast cancer: A prognostic study.

553 Background: Tumor-infiltrating lymphocytes (TILs) are pivotal in predicting responses to preoperative chemotherapy in triple-negative breast cancer. Yet, their clinical adoption as a diagnostic marker is limited by the challenge of achieving high accuracy with objective measures. This study introduces a novel AI-based method to quantitatively assess stromal TILs within the tumor bed using hematoxylin and eosin (H&E)-stained slides, examining their correlation with disease-free survival (DFS) and overall survival (OS). 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. We then developed an AI-based model to detect the epithelium and lymphocytes (CD3/CD20) from H&E-stained slides. This model was trained using annotated datasets, which included 26,509 images of the epithelium and 12,273 images of lymphocytes. The tumor bed areas were precisely defined by pathologists. Subsequently, the AI model identified epithelial and lymphocyte regions within these predefined areas. The TILs score was then calculated as the ratio of the lymphocyte area to the total stromal area within the tumor bed, excluding the epithelial areas identified by the AI model. Finally, we assessed the prognostic significance of the TILs score for DFS and OS through ROC analysis and log-rank tests. Results: ROC analysis identified an optimal TILs score threshold of 0.0042 for both DFS and OS, with an AUC of 0.743 (95% CI, 0.622-0.864) for DFS and 0.686 (95% CI, 0.559-0.812) for OS. This threshold significantly stratified patients into groups with distinct DFS and OS outcomes (P=0.003 and P=0.008, respectively). Conclusions: AI quantification of TILs offers a promising prognostic tool in triple-negative breast cancer, enhancing the objectivity and clinical utility of pathological analyses. Accurate TILs measurement could revolutionize prognostication and treatment planning.