Fully automatic HER2 tissue segmentation for interpretable HER2 scoring

Fully automatic HER2 tissue segmentation for interpretable HER2 scoring

Breast cancer is the most common cancer in women, with HER2 (human epidermal growth factor receptor 2) overexpression playing a critical role in regulating cell growth and division. HER2 status, assessed according to established scoring guidelines, offers important information for treatment selectio...

Full description

Saved in:
Bibliographic Details
Main Authors: Mathias Öttl, Jana Steenpass, Frauke Wilm, Jingna Qiu, Matthias Rübner, Corinna Lang-Schwarz, Cecilia Taverna, Francesca Tava, Arndt Hartmann, Hanna Huebner, Matthias W. Beckmann, Peter A. Fasching, Andreas Maier, Ramona Erber, Katharina Breininger
Format: Article
Language:English
Published: Elsevier 2025-04-01
Series:Journal of Pathology Informatics
Subjects:
HER2
Histopathology
Deep learning
Semantic segmentation
HER2 scoring
Online Access:http://www.sciencedirect.com/science/article/pii/S2153353925000203
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Breast cancer is the most common cancer in women, with HER2 (human epidermal growth factor receptor 2) overexpression playing a critical role in regulating cell growth and division. HER2 status, assessed according to established scoring guidelines, offers important information for treatment selection. However, the complexity of the task leads to variability in human rater assessments. In this work, we propose a fully automated, interpretable HER2 scoring pipeline based on pixel-level semantic segmentations, designed to align with clinical guidelines. Using polygon annotations, our method balances annotation effort with the ability to capture fine-grained details and larger structures, such as non-invasive tumor tissue.To enhance HER2 segmentation, we propose the use of a Wasserstein Dice loss to model class relationships, ensuring robust segmentation and HER2 scoring performance. Additionally, based on observations of pathologists' behavior in clinical practice, we propose a calibration step to the scoring rules, which positively impacts the accuracy and consistency of automated HER2 scoring. Our approach achieves an F1 score of 0.832 on HER2 scoring, demonstrating its effectiveness. This work establishes a potent segmentation pipeline that can be further leveraged to analyze HER2 expression in breast cancer tissue.
ISSN:2153-3539

Similar Items