ASN's Mission

ASN leads the fight to prevent, treat, and cure kidney diseases throughout the world by educating health professionals and scientists, advancing research and innovation, communicating new knowledge, and advocating for the highest quality care for patients.

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: FR-PO991

Deep Learning-Based Histopathological Assessment of Renal Tissue

Session Information

Category: Transplantation

  • 1901 Transplantation: Basic

Authors

  • Hermsen, Meyke, Radboudumc, Nijmegen, Netherlands
  • Den boer, Marjolijn, Radboudumc, Nijmegen, Netherlands
  • de Bel, Thomas, Radboudumc, Nijmegen, Netherlands
  • Kers, Jesper, Academic Medical Center, Amsterdam, Netherlands
  • Roelofs, Joris J., Academic Medical Center, Amsterdam, Netherlands
  • Stegall, Mark D., Mayo Clinic, Rochester, Minnesota, United States
  • Alexander, Mariam P., Mayo Clinic, Rochester, Minnesota, United States
  • Smith, Byron H., Mayo Clinic, Rochester, Minnesota, United States
  • Smeets, Bart, Radboudumc, Nijmegen, Netherlands
  • Hilbrands, Luuk, Radboudumc, Nijmegen, Netherlands
  • van der Laak, Jeroen A.W.M., Radboudumc, Nijmegen, Netherlands
Background

Quantitative measures are often used for histopathological assessment of renal tissue. We trained a convolutional neural network (CNN) for multi-class segmentation of digitized periodic acid-Schiff(PAS)-stained renal tissue sections.

Methods

The CNN was trained using annotations of 40 whole-slide images of PAS-stained renal transplant biopsies. Multi-class segmentation performance was assessed by calculating Dice coefficients (DCs) for 10 tissue classes on 10 transplant biopsies from Radboudumc and on 10 transplant biopsies from the Mayo clinic. Additionally, we fully segmented 15 nephrectomy samples and assessed the CNN’s glomerular detection rates. Lastly, CNN-based measures were compared with visually scored histological (Banff) components in 82 transplant biopsies.

Results

The weighted mean DCs were 0.80 and 0.84 in 10 transplant biopsies from Radboudumc and the Mayo Clinic, respectively. The ‘glomeruli’ class was best segmented in both data sets (DC 0.95 and 0.94), followed by ‘tubuli combined’ and ‘interstitium’. An example of the CNN’s visual output is shown in Figure 1. The CNN detected 92.7% of all glomeruli in nephrectomy samples, with 10.4% false positives. In whole transplant biopsies, the mean intraclass correlation coefficient for glomerular counting performed by pathologists and the CNN was 0.94. Moderate to strong correlations were observed between components of the Banff scoring system and CNN-based measures (Table 1).

Conclusion

This study presents the first CNN for multi-class segmentation of PAS-stained nephrectomy samples and transplant biopsies. Our CNN can be of aid for quantitative studies concerning renal histopathology across centers and provides opportunities for deep learning applications in routine diagnostics.

Figure 1. CNN segmentation result of a PAS-stained renal transplant biopsy.

Table 1. Mean Spearman’s ρ for visually scored (Banff) components and CNN-based measures in 82 transplant biopsies.

Funding

  • Government Support - Non-U.S.