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: TH-PO995

A Deep Learning Approach for Detection of Normal and Globally Sclerotic Glomeruli on Whole Slide Images from Renal Biopsy Sections Stained with H&E

Session Information

Category: Pathology and Lab Medicine

  • 1501 Pathology and Lab Medicine: Basic

Authors

  • Chen, Yijiang, Case Western Reserve University, Cleveland, Ohio, United States
  • Janowczyk, Andrew, Case Western Reserve University, Cleveland, Ohio, United States
  • Palmer, Matthew, University of Pennsylvania, Philadelphia, Pennsylvania, United States
  • Sedor, John R., Cleveland Clinic, Cleveland, Ohio, United States
  • O'Toole, John F., Cleveland Clinic, Cleveland, Ohio, United States
  • Barisoni, L., U. Miami, Miller School of Medicine, Miami Beach, Florida, United States
  • Madabhushi, Anant, Case Western Reserve University, Cleveland, Ohio, United States
Background

Clinically useful quantitative assessment of renal biopsies requires robust methodologies. We demonstrated that accurate estimates of glomerular global sclerosis (GS) adjusted for age can predict outcomes across proteinuric diseases. However, visual assessment is poorly reproducible and time consuming. To increase efficiency, we present a deep learning-based approach for detection and segmentation (object extraction) of normal and globally sclerotic glomeruli (histologic primitives) digital renal biopsies.

Methods

A convolutional neural network (CNN) architecture based on Alexnext was trained using transfer learning to produce 3 classifiers: (i) normal glomerular tuft, (ii) normal glomerulus (tuft + Bowman’s capsule), and (iii) GS. We used 74 cropped images from whole slide images of NEPTUNE renal biopsies sections stained with H&E, divided into training and testing sets.

Results

The detection and pixel level segmentation results for all 3 classifiers was graded via the F-score (A measure of classification accuracy that considers both precision and recall; 1 = perfect, 0 = worst).

Conclusion

The development of these 3 CNNs represents the first step in evaluating renal biopsies using novel machine-human interactive protocols. The classifier “normal glomerulus” is a better denominator compared to “normal glomerular tuft” to estimate percentage of GS. Improvement in F-score for the classifier GS, however, is necessary before automatic assessment of GS adjusted for age can be applied in clinical research and practice. Future steps to implement extended machine-human interactive protocols will include generation of other classifiers for the annotation of additional glomerular, tubulointerstitial and vascular normal and abnormal histologic primitives.

ModelMagnification# study subject (cropped images) for training# study subject (cropped images) for trainingSegmentation F-scoreDetection rate
Model
Normal glomerular tuft
5X27
(47 images)
14
(27 images)
0.8080.938
Normal glomerular tuft5X27
(47 images)
14
(27 images)
0.8440.985
Global glomerulosclerosis8X25
(53 images)
15
(35 images)
0.7000.836

Funding

  • NIDDK Support