Abstract: FR-OR013
Integrating Pathomic and Clinical Data Using Artificial Intelligence (AI) Models from Kidney Deceased Donors to Reduce Organ Discard and Improve Transplant Outcomes
Session Information
- Artificial Intelligence and Data Science Transforming Kidney Care: From Algorithms to Action
November 07, 2025 | Location: Room 361A, Convention Center
Abstract Time: 05:00 PM - 05:10 PM
Category: Artificial Intelligence, Digital Health, and Data Science
- 300 Artificial Intelligence, Digital Health, and Data Science
Authors
- Rodrigues, Luis, Universidade de Coimbra, Coimbra, Coimbra District, Portugal
- Paul, Anindya S., University of Florida, Gainesville, Florida, United States
- Rubin, Jeremy, University of Pennsylvania, Philadelphia, Pennsylvania, United States
- Abdelazim, Haitham M, University of Florida, Gainesville, Florida, United States
- Gupta, Akshita, University of Florida, Gainesville, Florida, United States
- Border, Samuel, University of Florida, Gainesville, Florida, United States
- Pardinhas, Clara, Universidade de Coimbra, Coimbra, Coimbra District, Portugal
- Sousa, Vitor Manuel Leitão de, Universidade de Coimbra, Coimbra, Coimbra District, Portugal
- Figueiredo, Arnaldo, Universidade de Coimbra, Coimbra, Coimbra District, Portugal
- Zee, Jarcy, University of Pennsylvania, Philadelphia, Pennsylvania, United States
- Sarder, Pinaki, University of Florida, Gainesville, Florida, United States
Background
The limited availability of donor kidneys remains a major barrier to transplantation. Procurement biopsies help assess organ quality but are linked to high discard rates, raising concerns that viable kidneys may be excluded due to current evaluation limitations. We developed ComPRePS [https://compreps.wiki/home], a cloud-based AI/ML computational pathology platform for kidney biopsy analysis. This study aimed to build a model combining pathological and clinical data to better assess graft potential and reduce discard rates.
Methods
We analyzed 211 procurement core biopsies from 152 deceased donors: 139 kidneys from 110 donors were transplanted, while 72 from 42 donors were discarded based on biopsy findings. Samples were FFPE-processed, PAS-stained, and digitized at 40X. Donors were 64.5% male, mean age 62.3 years; KDPI averaged 69.5 (39.5% >85%). Using ComPRePS, we segmented glomeruli, tubules, arteries, arterioles, and lesions (e.g., IFTA, vascular fibrointimal sclerosis). From these, 306 pathomic features were extracted, quantifying size, texture, and morphology across eosinophilic, luminal, and nuclear subregions.
We computed a hybrid pathologist–AI lesion score (Remuzzi: RZAI) and tested regression models. Features were selected using Minimum Redundancy Maximum Relevance, with bootstrapping and five-fold cross-validation. Model performance was evaluated using mean squared error and C-statistics. Two random forest models were selected: one to predict 12-month eGFR and another for Delayed Graft Function (DGF). We then estimated potential discard reduction by identifying discarded kidneys using thresholds: eGFR ≥30 (CKD stage 3), RZAI ≤5 (on-call pathologist cutoff for usability), and low DGF risk (Youden’s index <0.49). Projected organ utilization gains were calculated.
Results
Applying these criteria, discard reduction ranged from 56.9% to 100%, with projected utilization gains from 5% to 44.6% (Table 1).
Conclusion
Computational pathology significantly improves graft prediction and supports better donor kidney utilization.
Funding
- NIDDK Support