Pilot Evaluation of a Deep Learning Model for Nasogastric Tube Verification on Chest Radiographs: A Single-Center...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Accurate nasogastric (NG) tube placement confirmation is critical for patient safety.
• Current methods face delays and interpretation variability.
• Deep learning (DL) models offer potential but require real-world validation, especially for malpositioned tubes.

Purpose of the Study:

• To evaluate the real-world performance of a deep learning (DL) model for nasogastric (NG) tube confirmation.
• To assess the DL model's accuracy in detecting both correctly and incorrectly positioned NG tubes.
• To identify factors contributing to DL model misclassifications.

Main Methods:

• Pilot evaluation of a pre-existing DL model using 135 chest radiographs.
• Reference standard established by expert physicians.
• Performance assessment using ROC and PRC analyses, with statistical tests for misclassified cases.

Main Results:

• The DL model achieved high accuracy, correctly classifying 129 out of 135 cases.
• High sensitivity (96.1%) and PPV (99.2%) for correctly positioned tubes.
• Limited and unstable performance for detecting incomplete placements (NPV 54.5%).

Conclusions:

• The DL model is effective for confirming correctly placed NG tubes but requires improvement for malpositioned tube detection.
• Physician oversight is mandatory due to safety implications of misclassification.
• Larger, multi-center studies are needed to enhance reliability and clinical implementation.