Abstract: SA-PO0012
Automated Detection and Scoring of Microvascular Inflammation in Kidney Allografts Using Multiclass Segmentation Models
Session Information
- Intelligent Imaging and Omics: Phenotyping and Risk Stratification
November 08, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Gupta, Akshita, University of Florida, Gainesville, Florida, United States
- Sarder, Pinaki, University of Florida, Gainesville, Florida, United States
- Afsari, Fatemeh, University of Florida, Gainesville, Florida, United States
- La Rosa, Patricio Salvatore, Bayer CropScience LLC, St. Louis, Missouri, United States
- Rosenberg, Avi Z., Johns Hopkins University, Baltimore, Maryland, United States
Background
Microvascular Inflammation (MVI) is a critical histological marker in kidney transplant biopsies, characterized by immune cell infiltration in peritubular capillaries (PTCs) and glomeruli. Accurate assessment of MVI is essential for diagnosing antibody-mediated rejection and guiding post-transplant management. However, current evaluation methods rely on pathologists' subjective visual semiquantitative scoring, subject to inter-observer variability and of limited predictive value of donor-specific antibody (DSA) strength.
Methods
We present an automated pipeline to analyze whole slide images (WSI) of dual CD45 (leukocyte) and endothelial (CD34) immunohistochemical (IHC) staining to identify and quantify MVI. Our approach employs a multicompartment TransUNet model for segmenting glomeruli and arteries, and a separate EfficientNet-based model for pixel-wise segmentation of PTCs and basic thresholding for inflammatory cells. The output masks are post-processed to compute Inflamed Peritubular Capillary Ratio (iptcr), a quantitative surrogate of PTC inflammation which is highly predictive of DSA levels.
Results
Our models demonstrated high segmentation performance across tissue compartments (dice scores: 90% for glomeruli and 85% for arteries and PTCs). The computed iptcr score was highly correlated with manual iPTCT score (R2 0.83) . We have also developed a cloud-based tool for everyone to access and use.
Conclusion
This automated, reproducible method enables objective scoring of MVI in kidney allograft biopsies. Reducing reliance on semiquantitative scoring, facilitates faster and more consistent clinical decision-making with predictive capacity for the strength of clinically relevant DSA. Integration into digital pathology workflows could significantly enhance transplant diagnostics and patient care.
Funding
- NIDDK Support