Artificial Intelligence for the Verification of Nasogastric Tube Position in Chest Radiographs: A Systematic Review

Abstract

A systematic review examining the use of artificial intelligence (AI) to assess nasogastric tube (NGT) position on chest radiographs (CXR). Incorrect NGT placement can result in serious complications, including pulmonary feeding and aspiration. Chest radiography is commonly used to confirm tube position, but interpretation and reporting times may vary, particularly outside normal working hours. AI-based systems have therefore been developed to assist with tube detection, localisation and assessment of correct or incorrect placement. The purpose of this review is to evaluate the diagnostic performance of these systems, determine how well performance generalises beyond the datasets and institutions in which models were developed, and assess the strength of the current evidence for clinical implementation. The review includes original studies evaluating AI, machine learning or deep learning methods for assessment of NGT position on chest radiographs. Studies assessing closely related nasoenteric or enteric feeding-tube positioning tasks are also considered where findings are directly applicable to NGT safety. Mixed-device studies are included only where NGT-specific positioning outcomes can be separately extracted. Studies assessing tube presence alone, non-radiographic methods, non-AI approaches, or mixed-device performance without extractable NGT positioning outcomes are excluded. Particular attention is given to clinically relevant performance, including false-negative results, the number and proportion of malpositioned tubes within evaluation datasets, external and multicentre validation, and changes in performance across institutions or patient populations. Class imbalance, localisation methods, domain shift and reuse of public datasets or related model lineages are also examined, as these factors may affect the independence and generalisability of the available evidence. The intended outcome is a structured assessment of the current evidence for AI-assisted NGT position verification on chest radiographs, identifying both areas of promise and limitations relevant to clinical implementation. The review will also identify priorities for future research, including prospective multicentre validation, representative populations and malposition prevalence, independent datasets, and explicit assessment of clinically important missed unsafe tubes.