Tumour region identification guided scoring (TRIGS) and foundation model-based Tumour Infiltrating Lymphocyte scoring are prognostic for pathological complete response/event free survival in the triple negative patients in the PARTNER randomized controlled trial
Authors
Affiliations (1)
Affiliations (1)
- Dept of Histopathology, Cambridge University Hospitals NHS Foundation Trust, Cambridge United Kingdom
Abstract
Assessment of tumour infiltrating lymphocytes (TILs) is a robust prognostic biomarker for HER2-positive and triple negative breast cancer. We aimed to update a previously established pipeline for automated TIL assessment to align to clinical scoring guidelines (tumour region identification guided scoring), foundation model-based lymphocyte detection (SAM-TIL) and compare with alternative methods (HoVerNet, muTILs) and gold standard clinical assessment. TRIGS and SAM-TIL had an odds ratio of 1.95 (95% confidence interval(ci): 1.22-3.03, p=0.005) and 2.32 (95% ci:1.43-3.77, p=0.001) for predicting pathological complete response (pCR) rate in 166 neoadjuvantly treated patients in the TransNEO cohort. Hazard ratios for overall survival in 277 triple negative and HER2-positive patients in The Cancer Genome Atlas were 0.79 (95% ci: 0.63-1.00, p=0.05) and 0.80 (95% ci: 0.67-0.97, p=0.02) for TRIGS and SAM-TIL. Comparator methods showed similar results. Correlation with gold standard clinical assessment in 285 patients from the PARTNER randomized trial ranged from 0.59-0.69, which is substantially more than interobserver variability for the gold standard. Despite differences with gold standard assessment, no substantial difference in predicting pCR (AUC 0.60-0.64) or event free survival (Integrated Brier score approximately 0.10) were observed between the tested methods and gold standard assessment, suggesting AI tools that do not follow manual scoring guidelines could be validated and subsequently used. Although we reached prognostic performance similar to published literature and gold standard assessment, the lymphocyte detections produced by the models are not interchangeable with gold standard clinical assessment (correlation 0.59-0.69) and therefore cannot support pathologist assessment. Further studies to validate independent use and/or to improve prognostication are required.