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 X

Kidney Week

Abstract: SA-PO1076

Deep Learning for Segmentation of Glomeruli, Interstitial Fibrosis, and Tubular Atrophy in Renal Biopsies

Session Information

Category: Pathology and Lab Medicine

  • 1502 Pathology and Lab Medicine: Clinical

Authors

  • Jen, Kuang-Yu, University of California, Davis, Sacramento, California, United States
  • Ginley, Brandon G., SUNY Buffalo, Buffalo, New York, United States
  • Lutnick, Brendon, SUNY Buffalo, Buffalo, New York, United States
  • Tomaszewski, John E., University at Buffalo, Buffalo, New York, United States
  • Sarder, Pinaki, SUNY Buffalo, Buffalo, New York, United States
Background

Evaluation of the renal biopsy typically involves assessment of the renal cortical compartments, which consist of glomeruli, tubules, interstitium, and vessels. Of particular importance is the degree of what has been considered “chronic” or irreversible injury in the form of glomerulosclerosis, interstitial fibrosis and tubular atrophy (IFTA). These features are the main indicators for renal prognosis, irrespective of the etiology of renal disease. However, evaluation for such features can be imprecise and variable between experts, especially for IFTA. We have tested a deep convolutional neural network (CNN) segmentation of IFTA and glomeruli from whole slide images (WSIs) in a limited dataset of diabetic nephropathy (DN) biopsies.

Methods

Seven biopsies of patients diagnosed with DN were used. One WSI from each patient was completely annotated for IFTA and glomeruli. A Python interface was designed to accept WSI annotations from Aperio ImageScope, train a deep CNN, and perform subsequent WSI predictions. The network selected was Deeplab V2, implemented in Tensorflow.

Results

In this first study, training and testing was performed on the same 7 biopsies, due to data limitations. The deep CNN achieved significant performance on 3 out of the 7 biopsies tested, for both IFTA and glomeruli. For these cases the network achieved specificity for segmenting both IFTA and glomeruli (see Fig. 1).

Conclusion

For the first time we have shown computational segmentation of IFTA and glomeruli from renal biopsies. The Deeplab V2 system demonstrates the capability for high performance in addition to generalization to other histological compartments with an increase in dataset size. Future work will increase the training cases and implement a validation set.

Figure 1. Segmentation of IFTA and glomeruli by deep CNN. Green outlines regions of IFTA, yellow outlines glomeruli. Scalebar is 300µm.

Funding

  • NIDDK Support