Artificial intelligence in chronic kidney disease, dialysis and kidney transplantation: Current evidence, clinical applications, implementation challenges, and future directions

Abstract

Chronic kidney disease (CKD), dialysis-related complications, and kidney transplantation continue to impose a substantial global burden on morbidity, mortality, and healthcare utilization. In parallel, artificial intelligence (AI) has emerged as a potentially transformative tool for risk prediction, diagnostic support, treatment optimization, and longitudinal outcome assessment in nephrology. This review summarizes current evidence regarding AI applications in the early prediction of CKD and acute kidney injury, image-based diagnosis, dialysis decision support, mortality prediction in end-stage kidney disease and kidney transplantation, while also highlighting current limitations, clinical implementation challenges, and future directions for the responsible integration of AI into nephrology care. Available evidence suggests that machine-learning and deep-learning models can achieve promising predictive performance for CKD progression, acute kidney injury, and intradialytic complications. Convolutional neural network-based approaches have shown potential in kidney imaging, segmentation, and structural abnormality detection. Similarly, AI-based prognostic tools for dialysis and transplantation have demonstrated encouraging initial performance in estimating mortality, graft outcomes, and post-transplant complications. However, most published models remain limited by retrospective development, insufficient external validation, inadequate prospective testing, and substantial barriers to implementation, including dataset bias, limited interpretability, regulatory uncertainty, and workflow integration challenges. Overall, AI should be regarded as an adjunctive tool in modern nephrology rather than a replacement for clinical judgment. Robust validation, transparent reporting, and careful integration into healthcare systems will be required before routine clinical adoption can be justified.