Artificial intelligence for the monitoring of vascular access in hemodialysis: A narrative review focused on arteriovenous fistulas and grafts

Arteriovenous fistulas (AVFs) are the preferred vascular access for hemodialysis, yet they are affected by high rates of non-maturation and stenosis-related failure. Conventional surveillance methods—physical examination, duplex ultrasound, and angiography—are limited by operator dependence, cost, or invasiveness. Artificial intelligence (AI) has recently been applied to support non-invasive and semi-invasive monitoring of arteriovenous access. A narrative literature review was conducted using PubMed and related sources (PubMed Central, Nature Digital Medicine, Scientific Reports), searching combinations of the terms “artificial intelligence,” “machine learning,” “deep learning,” “arteriovenous fistula,” “arteriovenous graft,” “vascular access,” “hemodialysis,” “stenosis,” and “maturation.” Two main research streams emerged. The first applies machine learning and deep learning to acoustic flow-sound analysis for early stenosis detection, with reported sensitivity and specificity frequently exceeding 0.90 in single-center studies. The second applies machine learning to clinical and duplex ultrasound parameters to predict AVF maturation and long-term patency, with growing interest in multicenter cohorts. Home-based and wearable AI-enabled monitoring systems (AIoT) are emerging as a complementary approach for continuous surveillance. AI-based tools show promising diagnostic performance for vascular access surveillance but remain limited by small sample sizes, lack of external validation, heterogeneous reference standards, and limited model interpretability. Larger prospective multicenter studies are needed before these tools can be integrated into routine clinical pathways.