The Role of Artificial Intelligence in Predicting Acute Kidney Injury: A Systematic Review of Machine Learning and Deep Learning Approaches

Introduction: Acute kidney injury (AKI) is a common and potentially reversible condition in hospital settings and is associated with adverse outcomes, such as increased mortality, prolonged hospitalization, and higher healthcare costs. In this context, artificial intelligence (AI) algorithms have emerged as promising approaches for the early identification of patients at risk of AKI. This study aimed to systematically review research that has applied AI models for AKI prediction, with a focus on their performance, interpretability, and clinical evaluation. Methods: A systematic search was conducted in PubMed, Scopus, and Web of Science databases. Studies were included if they applied machine learning or deep learning algorithms to predict AKI. Extracted data included the type of algorithm, sample size, number of features, performance metrics, degree of interpretability, and presence of clinical evaluation. The models were categorized into white-box and black-box algorithms. Results: Among the 208 included studies, most reported an AUROC greater than 0.90. AUROC values reached up to 0.94 for white-box models and up to 0.97 for black-box models. Regression analysis indicated that dataset size had a significant impact on the performance of white-box models, whereas no such relationship was observed in black-box algorithms. Many studies also reported accuracy values ranging from 75% to 95%. Conclusions: Despite their higher interpretability, white-box models were used less frequently. Enhancing transparency, increasing clinical expert involvement, and conducting real-world clinical evaluations are essential for the broader adoption and practical implementation of AI-based models in AKI prediction.