Artificial Intelligence in Kidney Transplant Pathology: Current Evidence, Limitations, and Relevance to the Banff Framework

Kidney International Reports, vol. 11, pp. 106614

Abstract

The Banff Classification, established in 1991, provides a global standard for diagnosing and grading kidney transplant pathology, evolving through regular consensus meetings to incorporate new advances. Initially designed to address the lack of uniformity in renal allograft biopsy reporting, the Banff system now integrates semiquantitative scores for acute and chronic lesions. Recent progress in artificial intelligence (AI) and deep learning (DL) has accelerated quantitative analysis in kidney histology, particularly for interstitial fibrosis and tubular atrophy (IFTA) and inflammation, reducing interobserver variability and supporting prognostic assessment. Targeted algorithms such as positive pixel count (PPC) and DL-based segmentation models have demonstrated robust performance in identifying renal compartments and quantifying injury; and AI-assisted approaches for vascular lesions and immune cell profiling show promise but are limited in data diversity, stain standardization, and external validation. Additional metrics, such as nephron size, IFTA foci density, and mesangial expansion (ME), have been explored, with computational metrics and AI models showing potential for improved reproducibility and clinical relevance. Despite concerns about the implications of AI in pathology, optimism prevails regarding its ability to augment human expertise and enhance diagnostic precision. Ongoing work within the Banff Digital Pathology Working Group aims to extend automated assessment to additional Banff lesions, including glomerulitis, peritubular capillaritis, arteritis, and tubulitis. As these tools mature, they may enable more comprehensive quantitative characterization of transplant biopsies and facilitate exploration of histopathologic patterns that have been difficult to study. The Banff system's iterative evolution and embrace of AI-driven digital pathology underscore its enduring impact and promise in kidney transplant medicine.

Authors 5

1. Aleksandar Đenić

   Mayo Clinic

   Affiliation as printed

   Division of Nephrology and Hypertension, Mayo Clinic, Rochester, Minnesota, USA

2. Dominique van Midden

   Radboud University Medical Center

   Affiliation as printed

   Department of Pathology, Radboud University Medical Center, Nijmegen, the Netherlands

3. Peter Boor Aachen Institute of Pathology, University Clinic Aachen

   RWTH Aachen University · Universitätsklinikum Aachen

   Affiliation as printed

   Institute of Pathology, University Clinic Aachen, RWTH University, Aachen, Germany

4. Alton B. Farris

   Emory University

   Affiliation as printed

   Department of Pathology and Laboratory Medicine, Emory University, Atlanta, Georgia, USA

5. Kim Solez corresponding

   University of Alberta

   Affiliation as printed

   Department of Laboratory Medicine and Pathology, University of Alberta, Edmonton, Alberta, Canada