Abstract: PO1670
Deep Learning-Based Segmentation and Quantification of Podocyte Foot Process Morphology
Session Information
- Podocyte Injury in Human Disease: Pathomechanism, Diagnosis, and Therapy
November 04, 2021 | Location: On-Demand, Virtual Only
Abstract Time: 10:00 AM - 12:00 PM
Category: Glomerular Diseases
- 1204 Podocyte Biology
Authors
- Butt, Linus, Uniklinik Koln, Koln, Nordrhein-Westfalen, Germany
- Unnersjö-Jess, David, Uniklinik Koln, Koln, Nordrhein-Westfalen, Germany
- Höhne, Martin, Uniklinik Koln, Koln, Nordrhein-Westfalen, Germany
- Sergei, German, University of Cologne Center for Molecular Medicine Cologne, Cologne, Nordrhein-Westfalen, Germany
- Wernerson, Annika, Karolinska Institutet Enheten for medicinska njursjukdomar, Huddinge, Stockholm, Sweden
- Witasp, Anna, Karolinska Institutet Enheten for medicinska njursjukdomar, Huddinge, Stockholm, Sweden
- Schermer, Bernhard, Uniklinik Koln, Koln, Nordrhein-Westfalen, Germany
- Bozek, Kasia, University of Cologne Center for Molecular Medicine Cologne, Cologne, Nordrhein-Westfalen, Germany
- Benzing, Thomas, Uniklinik Koln, Koln, Nordrhein-Westfalen, Germany
Background
The advent of super-resolution light microscopy enabled imaging of the nanoscale dimensions of podocyte foot processes and the slit diaphragm and subsequent quantification of morphological alterations upon glomerular injury. However, these morphological analyses require manual work, which is time-consuming and investigator-dependent.
Methods
We used novel sample preparation protocols and applied super-resolution STED microscopy and conventional confocal microscopy to image podocyte foot processes in murine and human kidney tissue. Deep learning-based segmentation was utilized to automatically segment both the slit diaphragm pattern as well as several thousands of individual foot processes per sample.
Results
Our algorithm, the automatic morphological analysis of podocytes (AMAP), segmented the FPs and the SD at high accuracy and more effectively as compared to a previously published semi-automatic dataset. The morphological quantifications show a high agreement with our previous analysis, thereby confirming the correlation of albuminuria with certain morphological alterations of podocytes. In addition, we applied AMAP to human patient tissue and found different patterns of effacement in different disease entities.
Conclusion
The combination of three-dimensional optical imaging and deep-learning segmentation can be used to perform extensive morphological analyses of podocyte in health and disease. It confirms our previous semi-automatically performed analyses in a mouse model of FSGS and can be applied to patient material in order to assess morphological alterations in glomerular disease while eliminating investigator-bias. We believe AMAP can in the future complement the diagnostic algorithms in research and clinical pathology.
Funding
- Government Support – Non-U.S.