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: TH-PO802

Development and Validation of a Multilayer Segmentation Model to Quantify Chronic Changes on Kidney Biopsy

Session Information

Category: Pathology and Lab Medicine

  • 1800 Pathology and Lab Medicine

Authors

  • Asghar, Muhammad Sohaib, Mayo Clinic Minnesota, Rochester, Minnesota, United States
  • Stetzik, Lucas, Aiforia Technologies, Cambridge, Massachusetts, United States
  • Jagtap, Jaidip M., Mayo Clinic Minnesota, Rochester, Minnesota, United States
  • Barisoni, Laura, Duke University, Durham, North Carolina, United States
  • Alexander, Mariam P., Mayo Clinic Minnesota, Rochester, Minnesota, United States
  • Salem, Fadi E., Mayo Clinic in Florida, Jacksonville, Florida, United States
  • Smith, Maxwell L., Mayo Clinic Arizona, Scottsdale, Arizona, United States
  • Janowczyk, Andrew, Emory University, Atlanta, Georgia, United States
  • Mullan, Aidan F., Mayo Clinic Minnesota, Rochester, Minnesota, United States
  • Denic, Aleksandar, Mayo Clinic Minnesota, Rochester, Minnesota, United States
  • Rule, Andrew D., Mayo Clinic Minnesota, Rochester, Minnesota, United States
Background

Semiquantitative visual assessment of chronic structural changes is clinically relevant but has limited reproducibility. To better quantify chronicity, we developed a multilayer artificial intelligence (AI) model for the automated detection of glomerular, tubulointerstitial, and vascular chronic changes.

Methods

HistoQC was first applied to 1258 whole slide images (WSI) generated by multisite multi-scanner Aging Kidney Anatomy study to curate the dataset and to ensure representative variability. Aiforia Create, an AI-based interactive tool, was then used on 193 PAS-stained whole slide images (WSI) from 99 kidney tumor patients and 94 living kidney donor biopsies. This approach uses an iterative process to assign classes (semantic segmentation or object detection with instance segmentation) across nested layers. A total of 20,509 objects or regions were manually annotated to train the model. To validate the model, an independent test set of 10 WSIs (5 kidney tumor patients and 5 kidney donors) were annotated independently by 7 human validators and by the AI model. Then, for all 203 WSIs, the objects and regions detected by the AI model were used to calculate chronic changes. These were correlated with similar measures of chronic changes derived independently using human annotations.

Results

The final model assigned 20 classes across 9 nested layers (Figure 1). The F1 score for AI vs human was 95% and for human vs human was 96%. The correlation between AI model vs human measures was r=0.92 for glomerular volume, r=0.90 for cortex per glomerulus, r=0.92 for % globally sclerotic glomeruli, r=0.82 for AI % tubular atrophy vs. human % interstitial fibrosis/tubular atrophy, and r=0.72 for % artery stenosis from intimal thickening.

Conclusion

A multilayer AI model for the segmentation of 20 different classes on PAS-stained WSIs facilitate the quantification of kidney chronicity indices.

Figure 1. Layer tree of AI model with examples of performance in each class.

Funding

  • NIDDK Support