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-PO974

Automated Podocyte Foot Process Width Measurement Using Deep Learning

Session Information

Category: Pathology and Lab Medicine

  • 1601 Pathology and Lab Medicine: Basic

Authors

  • Najafian, Behzad, University of Washington, Seattle, Washington, United States
  • Sargolzaeiaval, Forough, University of Washington, Seattle, Washington, United States
  • Mauer, Michael, University of Minnesota, Minneapolis, Minnesota, United States
Background

Increased foot processes (FP) width (FPW) is an important measure of podocyte injury. There is no consensus on how to estimate average FPW. The current gold standard (unbiased stereology) is time consuming and not widely available. We aimed to automate average FPW estimation using deep learning (DL).

Methods

A custom multi-layered deep learning model was trained on a normalized electron-microscopy (EM) dataset of 800 images (augmented 10000+) obtained at ~40,000X. Images were captured using systematic uniform random sampling. Testing was done on 29 new kidney biopsies (30-157 images per biopsy) from patients with Fabry disease and variable proteinuria and pathology severity and a compiles set of images from kidney donors as normal controls. DL FPW measurements were compared with measurements done by an experienced technician using unbiased stereology. Measurements were correlated with available clinical and structural parameters.

Results

The automated report utility substantially reduced the time needed for average FPW measurement per biopsy (<1 min DL vs. 6-8 hours human). The DL model accuracy based on human segmentation as the ground truth on a scale of 0-1 (1=perfect) accuracy was 0.8 for glomerular basement membrane and 0.6 for slits. DL measurements (737±151nm) were ~6.5% smaller (p=0.03) than human measurements (788±194nm), but these two were correlated (r=0.77; p<0.0001). Bland-Altman plot showed that ~94% of DL vs. human differences fell within mean ± 1.96 SD of the differences. Both human and DL showed increased FPW in Fabry patients compared with controls and similar correlations between age and FPW in Fabry patients. Globotriaosylceramide inclusion density in podocytes correlated with DL-measured FPW (r=0.33, p=0.04), but human measurements did not. DL-measured FPW showed a trend with urine protein excretion rate (r=0.31, p=0.058), but human measurements did not. DL showed substantial variability in individual FPW in Fabry patients but not in controls.

Conclusion

DL algorithms while substantially reduced time needed for FPW measurement, provided reasonably accurate data correlating with human stereology measurements and with relevant clinical correlations. In addition, DL readily provided additional information on individual FPW variability which may be useful in podocyte injury assessment in secondary podocytopathies.

Funding

  • Other NIH Support