Breast cancer patient stratification using domain adaptation based lymphocyte detection in HER2 stained tissue sections

Ansh Kapil, Armin Meier, Anatoliy Shumilov, Susanne Haneder, Helen Angell, Günter Schmidt

Keywords: Computational Pathology, Deep Learning, Domain Adaptation, Breast Cancer

TL;DR: Domain adaptation based lymphocyte detection - Annotation transfer from H&E to HER2 IHC images and prognostic value of TILs in breast cancer.

Abstract: We extend the CycleGAN architecture with a style-based generator and show the efficacy of the proposed domain adaptation-based method between two histopathology image domains - Hematoxylin and Eosin (H&E) and HER2 immunohistochemically (IHC) images. Using the proposed method, we re-used large set of pre-existing annotations for detection of tumor infiltrating lymphocytes (TILs), which were originally done on H&E, towards a TIL detector applicable on HER2 IHC images. We provide analytical validation of the resulting TIL detector. Furthermore, we show that the detected stromal TIL densities are significantly prognostic as a biomarker for patient stratification on a triple-negative breast cancer (TNBC) cohort.

Paper Type: both

Primary Subject Area: Detection and Diagnosis

Secondary Subject Area: Transfer Learning and Domain Adaptation

Paper Status: original work, not submitted yet

Source Code Url: The source code is currently hosted in a closed organization repository and can be made available on request to authors.

Data Set Url: The dataset used in this study involves sensitive patient data and can not be shared/published

Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.

Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.