Establishing a functional arteriovenous fistula (AVF) is crucial to ensure effective dialysis treatment. However, there are currently no definitive criteria for predicting AVF maturation using preoperative ultrasound. This study aims to assess the efficacy of AI-driven models in analyzing comprehensive ultrasonographic variables across multiple forearm locations to predict successful AVF maturation. A retrospective analysis was conducted on 492 cases within a cohort of 494 individuals who underwent radiocephalic arteriovenous fistula (AVF) creation across three centers over a span of four years. The analysis involved various variables such as demographic data, cephalic vein diameters at distal (vein 1), mid (vein 2), and proximal (vein 3) locations, vein wall thickness, competing tributaries, peak systolic velocity, and radial artery waveform. To assess the predictive value of these variables, XGBoost, a machine learning model, was utilized. Significant predictors of AVF maturation included cephalic vein diameters (distal, mid, and proximal) with p-values of 0.004, 0.003, and 0.009, respectively, and competing tributaries (p < 0.001). The optimal model performance was in Stage 4, excluding vein 1 data, achieving an F1 score of 0.883, a Matthews Correlation Coefficient of 0.741, and an AUC of 86.9%. A vein 2 diameter cut-off of 4.5 mm resulted in an 89.5% success rate in maturation. This study underscores the value of comprehensive ultrasonographic assessments in predicting AVF maturation, advocating for detailed preoperative ultrasound mapping of the entire forearm to enhance AVF outcomes.
Cho et al. (2026) studied this question.