Improved HER2 Tumor Segmentation with Subtype Balancing Using Deep Generative Networks

Öttl M, Steenpaß J, Rübner M, Geppert CI, Qiu J, Wilm F, Hartmann A, Beckmann M, Fasching P, Maier A, Erber R, Breininger K (2023)

Publication Type: Conference contribution

Publication year: 2023

Journal

Proceedings - International Symposium on Biomedical Imaging

Publisher: IEEE Computer Society

Book Volume: 2023-April

Conference Proceedings Title: Proceedings - International Symposium on Biomedical Imaging

Event location: Cartagena CO

ISBN: 9781665473583

DOI: 10.1109/ISBI53787.2023.10230503

Abstract

Tumor segmentation in histopathology images is often complicated by its composition of different histological subtypes and class imbalance. Oversampling subtypes with low prevalence features is not a satisfactory solution since it eventually leads to overfitting. We propose to create synthetic images with semantically-conditioned deep generative networks and to combine subtype-balanced synthetic images with the original dataset to achieve better segmentation performance. We show the suitability of Generative Adversarial Networks (GANs) and especially diffusion models to create realistic images based on subtype-conditioning for the use case of HER2-stained histopathology. Additionally, we show the capability of diffusion models to conditionally inpaint HER2 tumor areas with modified subtypes. Combining the original dataset with the same amount of diffusion-generated images increased the tumor Dice score from 0.833 to 0.854 and almost halved the variance between the HER2 subtype recalls. These results create the basis for more reliable automatic HER2 analysis with lower performance variance between individual HER2 subtypes.

Authors with CRIS profile

Mathias Öttl Lehrstuhl für Informatik 5 (Mustererkennung) Jana Steenpaß Lehrstuhl für Allgemeine Pathologie und Pathologische Anatomie Matthias Rübner Department of Obstetrics and Gynaecology Carol-Immanuel Geppert Institute of Pathology Jingna Qiu Juniorprofessur für Artificial Intelligence in Medical Imaging Frauke Wilm Lehrstuhl für Informatik 5 (Mustererkennung) Arndt Hartmann Lehrstuhl für Allgemeine Pathologie und Pathologische Anatomie Matthias Beckmann Lehrstuhl für Geburtshilfe und Frauenheilkunde Peter Fasching Professur für Translationale Frauenheilkunde und Geburtshilfe Andreas Maier Lehrstuhl für Informatik 5 (Mustererkennung) Ramona Erber Institute of Pathology Katharina Breininger Juniorprofessur für Artificial Intelligence in Medical Imaging

How to cite

APA:

Öttl, M., Steenpaß, J., Rübner, M., Geppert, C.-I., Qiu, J., Wilm, F.,... Breininger, K. (2023). Improved HER2 Tumor Segmentation with Subtype Balancing Using Deep Generative Networks. In Proceedings - International Symposium on Biomedical Imaging. Cartagena, CO: IEEE Computer Society.

MLA:

Öttl, Mathias, et al. "Improved HER2 Tumor Segmentation with Subtype Balancing Using Deep Generative Networks." Proceedings of the 20th IEEE International Symposium on Biomedical Imaging, ISBI 2023, Cartagena IEEE Computer Society, 2023.

BibTeX: Download