AI-based tools for predicting and early diagnosis of graft rejection in solid organ transplantation - a systematic review.

Authors

Douglas Henderson, Tobias Niederegger, Robert Munzinger, Emre Karakas, Jonas Hildinger, Jasper Iske, Max Heiland, Curtis L Cetrulo, Alexandre G Lellouch, Leonard Knoedler

Published in

Transplantation reviews (Orlando, Fla.). Volume 40. Issue 4. Pages 101054. Aug 28, 2026. Epub Aug 28, 2026.

Abstract

Solid organ transplantation (SOT) is the standard therapeutic approach to end-stage organ failure. Artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), has emerged as a promising tool for analyzing large, complex datasets, enabling both prediction of rejection risk and early detection of established graft injury. This systematic review synthesizes current evidence on AI-based approaches for predicting future rejection risk and detecting active rejection in human SOT, evaluates their performance, and identifies gaps for future research.
This review followed PRISMA 2020 guidelines. PubMed/MEDLINE, EMBASE, and Web of Science were searched up to April 30, 2025, using terms related to AI and graft rejection. Eligible studies included peer-reviewed original research using AI to predict, detect, or monitor rejection in humans. Three reviewers independently screened titles, abstracts, and full texts, resolving disagreements by consensus. Due to heterogeneity in methods and objectives, meta-analysis was not feasible.
Of 195 studies identified, 62 met inclusion criteria. Most focused on kidney transplantation (n = 49, 79%), followed by heart (n = 6, 10%), liver (n = 4, 6%), lung (n = 1, 2%), and pancreas (n = 1, 2%). One study addressed multiple organs. Among diagnostic studies, AI, particularly ML and DL, demonstrated high diagnostic performance in non-kidney transplantation, often exceeding reported AUC of 0.90. In kidney transplantation, DL models, including convolutional neural networks and transformer-based architectures, reached accuracies up to 99.89% and AUCs up to 0.99. ML methods such as XGBoost, Bayesian classifiers, and logistic regression also performed well, with XGBoost achieving AUCs of 0.95-0.97, Bayesian classifiers reaching accuracies of 93.3% to 97%, and logistic regression models reporting AUC values up to 0.969. Among predictive studies, ML-based models similarly demonstrated strong discriminative performance.
AI models using ML and DL may show strong potential, particularly in kidney transplantation, for non-invasive early detection of active graft rejection and prediction of future rejection risk, across diverse data types. The exceptionally high performance reported by some studies warrants careful interpretation. Challenges such as lack of standardization, limited validation, and interpretability must be addressed through well-designed multicentre studies to support clinical translation.

PMID:
42673681
Bibliographic data and abstract were imported from PubMed on 01 Sep 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.