Abstract: FR-PO957
Computational Characterization of Arteries/Arterioles in FSGS/Minimal Change Disease
Session Information
- Top Trainee Posters - 3
October 26, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:30 AM - 11:30 AM
Category: Pathology and Lab Medicine
- 1800 Pathology and Lab Medicine
Authors
- Zhou, Jin, Duke University, Durham, North Carolina, United States
- Demeke, Dawit S., University of Michigan, Ann Arbor, Michigan, United States
- Li, Xiang, Duke University, Durham, North Carolina, United States
- Dinh, Timothy A., University of Michigan, Ann Arbor, Michigan, United States
- O'Connor, Christopher Lund, University of Michigan, Ann Arbor, Michigan, United States
- Liu, Qian, The Children's Hospital of Philadelphia, Philadelphia, Pennsylvania, United States
- Zee, Jarcy, University of Pennsylvania Perelman School of Medicine, Philadelphia, Pennsylvania, United States
- Chen, Yijiang, Stanford University, Stanford, California, United States
- Janowczyk, Andrew, Emory University, Atlanta, Georgia, United States
- Holzman, Lawrence B., University of Pennsylvania, Philadelphia, Pennsylvania, United States
- Mariani, Laura H., University of Michigan, Ann Arbor, Michigan, United States
- Bitzer, Markus, University of Michigan, Ann Arbor, Michigan, United States
- Barisoni, Laura, Duke University, Durham, North Carolina, United States
- Hodgin, Jeffrey B., University of Michigan, Ann Arbor, Michigan, United States
- Lafata, Kyle Jon, Duke University, Durham, North Carolina, United States
Background
The clinically used semi-qualitative scoring of arteriosclerosis is poorly standardized and reproducible. We developed a computational pipeline to precisely and reproducibly characterize the morphology of arteries/arterioles using digital kidney biopsies.
Methods
One trichrome-stained whole slide image (WSI) from N=229 NEPTUNE/CureGN participants with focal segmental glomerulosclerosis (N=128) or minimal change disease (N=101) was used. We developed, applied, and quality controlled deep learning models for segmentation of muscular vessels and intra-vascular compartments (lumen, intima, media). Each vessel was classified into (i) arcuate and (ii) interlobular artery, (iii) 2 muscle layers arteriole, and (iv) 1 muscle layer arteriole and venule. (i), (ii), and (iii) were visually scored for arteriosclerosis (0-3). Intra-arterial thickness was measured using radial sampling and ray casting, with average intima-media thickness ratio computed for each artery/arteriole for correlation with visual scores. These features were summarized for each WSI using aggregation metrics (median, 75th percentiles) for association with disease progression (40% eGFR decline or renal failure).
Results
N=1509 arterioles, 695 interlobular and 131 arcuate arteries were segmented. There was a statistically significant correlation between arteriosclerosis scores and intima-media thickness ratio average (Spearman ρ=0.28, p<0.0001 for arterioles; ρ=0.67, p<0.0001 for interlobular arteries; ρ=0.70, p<0.0001 for arcuate arteries). Arteriole features aggregated at median and 75th percentiles associated with disease progression (log-rank p<0.1). No association was identified in arcuate and interlobular artery features (log-rank p>0.1).
Conclusion
We developed a novel approach to automatically segment arteries/arterioles sub-compartments and to computationally characterize morphologic features of arteriosclerosis. This technique demonstrated potential as a promising reproducible tool to aid pathologists’ clinical assessment.
Funding
- NIDDK Support; Private Foundation Support