Kidney Week

Abstract: FR-OR55

Computationally Derived Tubular Features Are Prognostic of Clinical Outcomes in Glomerular Kidney Diseases

Session Information

• Kidney Pathology: From Classic Clinicopathologic Studies to Computational Pathology
  November 03, 2023 | Location: Room 109, Pennsylvania Convention Center
  Abstract Time: 05:06 PM - 05:15 PM

Category: Pathology and Lab Medicine

• 1800 Pathology and Lab Medicine

Authors

• Fan, Fan, Department of Biomedical Engineering, Case Western Reserve University, Cleveland, Ohio, United States

Fan Fan, MS

• Liu, Qian, The Children's Hospital of Philadelphia Research Institute, Philadelphia, Pennsylvania, United States

Qian Liu,

• Zee, Jarcy, The Children's Hospital of Philadelphia Research Institute, Philadelphia, Pennsylvania, United States

Jarcy Zee,

• Ozeki, Takaya, Department of of Internal Medicine, University of Michigan, Ann Arbor, Michigan, United States

Takaya Ozeki,

• Demeke, Dawit S., Department of Pathology, University of Michigan, Ann Arbor, Michigan, United States

Dawit S. Demeke,

• Wang, Bangchen, Department of Pathology, Duke University, Durham, North Carolina, United States

Bangchen Wang,

• Mariani, Laura H., Department of of Internal Medicine, University of Michigan, Ann Arbor, Michigan, United States

Laura H. Mariani,

• Lafata, Kyle Jon, Department of Radiation Oncology, Duke University, Durham, North Carolina, United States

Kyle Jon Lafata,

• Chen, Yijiang, Department of Biomedical Engineering, Case Western Reserve University, Cleveland, Ohio, United States

Yijiang Chen,

• Holzman, Lawrence B., Department of Medicine, Division of Nephrology and Hypertension, University of Pennsylvania, Philadelphia, Pennsylvania, United States

Lawrence B. Holzman,

• Hodgin, Jeffrey B., Department of Pathology, University of Michigan, Ann Arbor, Michigan, United States

Jeffrey B. Hodgin,

• Madabhushi, Anant, Department of Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, Georgia, United States

Anant Madabhushi,

• Barisoni, Laura, Department of Pathology, Duke University, Durham, North Carolina, United States

Laura Barisoni,

• Janowczyk, Andrew, Department of Biomedical Engineering, Emory University and Georgia Institute of Technology, Atlanta, United States

Andrew Janowczyk,

Background

The visual semiquantitative assessment of chronic and acute tubular damage has limited reproducibility and prognosticating/predictive power. We hypothesize that sophisticated, computationally assessed, domain-inspired tubular pathomic features can enhance prognostication of proteinuria diseases.

Methods

We developed and applied tubular segmentation algorithms for tubular lumen (TL), epithelium (TE), nuclei, and basement membranes (TBM), to 235 PAS NEPTUNE/CureGN PAS-stained whole slide images (124 FSGS,111 MCD/MCD-like). From these segmentations, 56 features were extracted and summarized at the patient level. We used MRMR to select 10 features most prognostic for time from biopsy to 40% eGFR decline/kidney failure, and proteinuria remission, and Ridge regression models to estimate prognostic accuracy.

Results

Features (Fig.2-A), reflecting TE simplification and TBM thickening/smoothing, were most prognostic of disease progression; when added to other parameters, they increased prognostic accuracy for both outcomes (Fig.2-B).

Conclusion

Computational quantification of tubular pathomic features provides prognostic information above routine measures in glomerular diseases.

2 selected features contrasting patients with worse/better outcome

Funding

• NIDDK Support