Correlation Between Artificial Intelligence and Clinical Expert Criteria in the Radiographic Evaluation of Tunneled Hemodialysis Catheters

Background: Artificial intelligence (AI) is rapidly emerging as a diagnostic aid across medical specialties. In interventional nephrology, accurate radiographic assessment of tunneled hemodialysis catheters is essential to ensure proper function and patient safety, especially in resource-limited settings. This study aims to evaluate the correlation between Expert Clinical Criteria (ECC) and AI-based assessment of catheter positioning. Methods: A cross-sectional observational study was conducted between February and April 2025 to evaluate 259 radiographs of newly tunneled hemodialysis catheters placed via the internal jugular vein. Each image was assessed by an Interventional Nephrologist using ECC, followed by AI-based classification using ChatGPT. A second AI evaluation was performed after implementing a rule-based algorithm reflecting ECC. Correlation between qualitative variables was analyzed using Spearman’s rho and Cohen’s kappa. The study took place at SENETO Clinic, Metepec, Mexico. Results: A total of 259 radiographs of tunneled hemodialysis catheters were analyzed. Based on ECC, 64% of catheters were classified as optimal, 25% as suboptimal, and 11% as inadequate. Initial evaluation by AI using ChatGPT showed moderate correlation with expert assessment, with a Spearman’s rho of 0.51 and a Cohen’s kappa of 0.48, indicating limited agreement beyond chance. After integrating a rule-based algorithm reflecting ECC parameters—specifically tip location and the presence or absence of tunneling angulation—the AI model demonstrated improved concordance with clinical judgment. Spearman’s rho increased to 0.78, and Cohen’s kappa to 0.71, indicating a substantial gain in correlation and agreement. These results were statistically significant (p < 0.05), supporting the potential utility of AI as a decision-support tool in the radiographic evaluation of hemodialysis catheter placement, especially when ECC are incorporated as the evaluative framework. Conclusion: AI demonstrates strong potential as a decision-support tool in the radiographic evaluation of hemodialysis catheters. However, its application should be approached with caution and always validated against expert clinical criteria to ensure patient safety and diagnostic accuracy. Funding: Private Foundation Support