Kidney Week

Abstract: SA-PO074

Applying Computer Vision to Evaluate Percent Global Glomerulosclerosis in Donor Kidney Biopsies

Session Information

• Transplantation: Recipient and Donor Assessment
  October 27, 2018 | Location: Exhibit Hall, San Diego Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Transplantation

• 1802 Transplantation: Clinical

Authors

• Kudose, Satoru, Washington University School of Medicine, St. Louis, Missouri, United States

Satoru Kudose,

• Hunt, Matthew S., Washington University School of Medicine, St. Louis, Missouri, United States

Matthew S. Hunt,

• Marsh, Jon N., Washington University School of Medicine, St. Louis, Missouri, United States

Jon N. Marsh,

• Swamidass, S. joshua J., Washington University School of Medicine, St. Louis, Missouri, United States

S. joshua J. Swamidass,

• Gaut, Joseph, Washington University School of Medicine, St. Louis, Missouri, United States

Joseph Gaut,

Background

Determining percent global glomerulosclerosis (GS) in donor kidney frozen wedge biopsies is used to determine organ viability. However, pathologic examination is labor-intensive and error prone. Previously, we developed a convolutional neural network (CNN) to quantify percent GS in scanned kidney biopsy whole slide images (WSI). The aim of this study is to compare CNN with pathologist evaluation of percent GS in scanned WSI.

Methods

Twenty consecutive donor kidney frozen wedge biopsies taken from April to August 2016 were scanned at 20x using an Aperio Scanscope CS scanner as part of routine clinical workflow. Four pathologists and one trainee manually enumerated globally sclerosed and non-sclerosed glomeruli using Aperio ImageScope, blinded to clinical data, in a setting mimicking routine clinical practice. The CNN evaluated the same WSI. Pathologist agreement was assessed using intraclass correlation coefficient (ICC).

Results

ICCs for pathologist evaluation of percent GS showed excellent agreement (0.8). Variability increased among cases with >10% GS. Percent GS determined by the CNN was within the pathologist range in 15/20 (75%) cases (Figure 1). CNN determined mean absolute difference of percent GS from pathologist median was 2% (range 0 – 6.1%).

Conclusion

The CNN quantifies percent GS in frozen section donor kidney wedge biopsies with performance comparable to pathologists and has potential to be incorporated in clinical practice.

Funding

• Private Foundation Support