Scoring Tumor-Infiltrating Lymphocytes in breast DCIS: A guideline-driven artificial intelligence approach

Matteo Pozzi, Natalie Klubickova, Michela Campora, Frederique Meeuwsen, Joey Spronck, Carlijn Lems, Michelle Stegeman, Leslie Tessier, Mattia Barbareschi, Jeroen van der Laak, Giuseppe Jurman, Francesco Ciompi

Keywords: DCIS, ductal carcinoma in situ, TILs score computation

TL;DR: Development of a AI assisted score to evaluate the abundance of tumor infiltrating lymphocytes in Ductal Carcinoma in Situ.

Abstract:

This study focuses on the assessment of Tumor-Infiltrating Lymphocytes (TILs) in Breast ductal carcinoma in situ (DCIS) by integrating artificial intelligence with international guidelines. DCIS is a non-invasive cancer with intrinsic potential to evolve to invasive breast cancer (IBC), making it critical to understand factors influencing this progression. TILs are a prognostic biomarker in IBC, but their role in DCIS remains under-explored. This work proposes an automated pipeline for computing TILs scores using deep learning for DCIS segmentation and TILs detection, following the guidelines of the International Immuno-Oncology Biomarker Working Group.
We report the inter-observer variability at TILs scoring among Pathologists and show that the AI-based TILs scores have good concordance with human assessments. Future research will aim to reduce false positives in DCIS segmentation and detection, support the reference standard with immunohistochemical staining, and expand the dataset to enhance the robustness of the TILs detection algorithm. Ultimately, this method aims to aid Pathologists in assessing the risk associated with DCIS lesions

Submission Number: 23