An Artificial Neural Network for Nasogastric Tube Position Decision Support

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Nasogastric tube (NGT) placement requires accurate verification on chest radiographs.
• Malpositioned NGTs pose risks, necessitating reliable detection methods.
• Junior physicians benefit from decision support tools for NGT safety assessments.

Purpose of the Study:

• Develop and validate a deep learning model for NGT malposition detection on chest radiographs.
• Assess the impact of AI as a clinical decision support tool for junior physicians.
• Improve the safety of feeding decisions in patients with NGTs.

Main Methods:

• A neural network ensemble was trained on over 1.1 million chest radiographs.
• The model was fine-tuned on 7,081 labeled chest radiographs and validated on 335 images.
• Junior physicians made feeding decisions with and without AI-generated malposition probabilities.

Main Results:

• The AI model achieved high accuracy in detecting satisfactory, malpositioned, and bronchial NGTs (AUCs 0.82-0.98).
• AI support significantly improved interreader agreement among junior physicians (0.65 to 0.77).
• AI-aided decisions increased agreement between junior physicians and radiologists (0.53 to 0.65).

Conclusions:

• A deep learning classifier for NGT malposition can aid junior physicians.
• AI-powered decision support enhances the accuracy of feeding safety assessments.
• This technology has the potential to improve patient safety in NGT management.