Abstract
Tumor-infiltrating lymphocytes (TILs) is a recognized prognostic biomarker in breast cancer. However, poor interobserver agreement and limited reproducibility highlight the need for computational approaches. Despite advances, adoption of computational models has been hindered by lack of standardized methods and robust benchmarks. To address this, we launched TIGER, an international competition to build open-source computational TILs (cTILs) models. Here, we present a multi-centric analysis of cTILs methods on resections and biopsies from 3,708 human epidermal growth factor receptor 2-positive (HER2+) or triple-negative breast cancers (TNBC) from clinical practice and phase 3 trials. We report benchmarks on image analysis performance, show strong agreement of cTILs with pathologists, and demonstrate positive association of cTILs with neoadjuvant therapy response in HER2+, superior to visually scored TILs. We also show that cTILs add independent prognostic information to clinical variables in TNBC resections. Data, methods and benchmarks are publicly available:
.