ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on Twitter

Kidney Week

Abstract: TH-PO565

Stain-Independent Segmentation and Quantification of Kidney Histopathology Using Deep Learning

Session Information

  • Pathology and Lab Medicine
    November 03, 2022 | Location: Exhibit Hall, Orange County Convention Center‚ West Building
    Abstract Time: 10:00 AM - 12:00 PM

Category: Pathology and Lab Medicine

  • 1700 Pathology and Lab Medicine

Authors

  • Boor, Peter, Universitatsklinikum Aachen, Aachen, Nordrhein-Westfalen, Germany
  • Hölscher, David Laurin, Universitatsklinikum Aachen, Aachen, Nordrhein-Westfalen, Germany
  • Klinkhammer, Barbara Mara, Universitatsklinikum Aachen, Aachen, Nordrhein-Westfalen, Germany
  • Bülow, Roman David, Universitatsklinikum Aachen, Aachen, Nordrhein-Westfalen, Germany
  • Lotz, Johannes, Fraunhofer-Institut fur Digitale Medizin MEVIS, Bremen, Bremen, Germany
  • Weiss, Nick, Fraunhofer-Institut fur Digitale Medizin MEVIS, Bremen, Bremen, Germany
  • Daniel, Christoph, Friedrich-Alexander-Universitat Erlangen-Nurnberg, Erlangen, Bayern, Germany
  • Amann, Kerstin U., Friedrich-Alexander-Universitat Erlangen-Nurnberg, Erlangen, Bayern, Germany
  • Bouteldja, Nassim, Universitatsklinikum Aachen, Aachen, Nordrhein-Westfalen, Germany
Background

In digital pathology, many image analysis tasks are challenged by the need for exhaustive and time-consuming data annotations to tackle various sources of variability in the image domain. Thus, there is a great need for methods addressing data variation without the manual overhead. E.g., in clinical routine and preclinical research, different tissue stainings are used that show differently colored and textured tissue. Deep learning (DL) models however are typically trained and thus limited to a single stain.

Methods

In this work, we deal with this inter-stain variation and propose a methodological concept based on stain augmentation to make DL models stain independent. We comprehensively compare this approach with two state-of-the-art methodologies including stain translation and image registration. Our aim was to make a pretrained and stain-specific DL segmentation model applicable to various other stains without the need for data annotations. We employed our previously published DL model that segments six major kidney structures from PAS-stained histopathology in different species.

Results

Validation on various immunohistochemical stainings in mice and humans showed that the proposed stain augmentation significantly outperformed both baseline approaches in all stains. It provided high segmentation accuracies in all structures and stains, even in held-out stains not seen during training. We used the stain-augmented model to facilitate compartment-specific morphometrical analysis of immunostainings in animal models and patient biopsies for the efficient analysis of inflammation and fibrosis.

Conclusion

This benchmark study suggests that stain augmentation is a highly effective approach to yield stain independence in DL models without manual overhead. This opens new possibilities for efficient and exhaustive computational immunohistomorphometry.

Funding

  • Government Support – Non-U.S.