Abstract: TH-PO0991
Artificial Intelligence (AI)-Based Quantification of Collagen Microarchitecture Across Serial Biopsies Reveals Longitudinal Remodeling Patterns After Kidney Transplantation
Session Information
- Transplantation: Basic - Immune Biology, Tissue Injury, and Emerging Technologies
October 22, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Transplantation
- 2001 Transplantation: Basic
Authors
- Subramanian, Ajay Rajaraman, Georgia Institute of Technology, Atlanta, Georgia, United States
- Tokuyama, Naoto, Emory University, Atlanta, Georgia, United States
- Abe, Yohei, Kagawa Daigaku, Takamatsu, Kagawa Prefecture, Japan
- Krishnamoorthy, Anusha, Georgia Institute of Technology, Atlanta, Georgia, United States
- Fan, Fan, Emory University, Atlanta, Georgia, United States
- Pathak, Tilak B., Emory University, Atlanta, Georgia, United States
- Taoka, Rikiya, Kagawa Daigaku, Takamatsu, Kagawa Prefecture, Japan
- Madabhushi, Anant, Emory University, Atlanta, Georgia, United States
Background
Early interstitial fibrosis has been associated with long-term graft function after kidney transplantation. However, collagen microarchitecture may undergo longitudinal fibrotic changes in the early post-transplant period, and the trajectory of these changes between implantation and 3 months remains uncharacterized. We investigated whether AI-based quantification of longitudinal collagen microarchitectural changes across serial biopsies could predict early renal allograft function decline.
Methods
We analyzed 136 living-donor kidney transplant recipients at Kagawa University. Masson trichrome–stained implantation and 3-month biopsies were analyzed using a computational pipeline extracting 24 collagen features capturing fiber structure, spatial organization, and textural heterogeneity. Collagen changes were evaluated using Wilcoxon signed-rank tests with Bonferroni correction. Patients were divided into training (n=91) and test (n=45) cohorts. A delta-change model was evaluated for prediction of 2-year ≥20% eGFR decline using Cox regression and Kaplan–Meier analysis.
Results
Eighteen features changed significantly after Bonferroni correction (p<0.002), indicating reduced collagen organization and increased structural heterogeneity. Two prognostic delta features were identified: Haralick IMC1 (HR=1.328, 95% CI 1.091–1.617, p=0.005) and collagen contrast (HR=1.367, 95% CI 1.036–1.804, p=0.027), both reflecting greater textural complexity and spatial irregularity of collagen deposition at 3 months. A two-feature model stratified the held-out test cohort into high- and low-risk groups, with high-risk patients showing over twice the rate of ≥20% eGFR decline events (log-rank p=0.020, C-index=0.637).
Conclusion
Changes in collagen microarchitecture between implantation and 3-month biopsies are associated with subsequent renal allograft dysfunction, suggesting that the interval change between serial biopsies may provide useful prognostic information.