AI-derived Tumor-Infiltrating Lymphocytes Enhance the Prediction of Pathologic Complete Response in Early-Stage Triple-Negative Breast Cancer

Stromal tumor-infiltrating lymphocytes (sTILs) may hold clinical value for triple-negative breast cancer (TNBC) for outcome prediction. Manual scoring of sTILs suffers from inter-observer variability, limiting its clinical adoption. We developed an artificial intelligence (AI) workflow to automatically quantify sTILs from diagnostic biopsies and evaluated its predictive performance for pathological complete response (pCR) in patients with early-stage TNBC receiving neoadjuvant chemotherapy. The pipeline combined tissue segmentation with context-aware cell classification to derive sTILs scoring, AI-sTILdensity. Using pretreatment biopsies (n = 394) from the ARTEMIS trial (ClinicalTrials.gov: NCT02276443; October 21, 2014), AI-sTILdensity showed strong correlation with manual sTILs (Spearman rho: 0.686-0.739) and improved prediction of pCR (AUC: 0.702-0.747 vs. 0.692-0.713) as an independent feature (P = 0.016). Predictive performance was consistent in an external TNBC set (n = 64). These findings demonstrated that automated assessment of sTILs from biopsy specimens is reproducible and can modestly enhance pCR prediction, supporting the use of AI approaches to enable more objective and scalable evaluation of sTILs in TNBC.