Detecting Endotracheal Tube and Carina on Portable Supine Chest Radiographs Using One-Stage Detector with a...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Critical Care Medicine

Background:

• Accurate endotracheal tube (ETT) placement is crucial in intensive care units (ICUs) to prevent complications.
• Current methods for verifying ETT position, often relying on chest radiographs, face challenges like image occlusion and patient positioning.
• Existing automated approaches frequently require manual intervention, limiting their clinical utility.

Purpose of the Study:

• To develop an automated system for precise localization of the ETT tip and Carina.
• To enable accurate detection of ETT malposition without manual intervention.
• To improve patient safety by enhancing the reliability of ETT position verification.

Main Methods:

• A novel deep learning architecture integrating FCOS (Fully Convolutional One-Stage Object Detection) with a Coarse-to-Fine Attention (CTFA) mechanism and a segmentation branch.
• Implementation of a post-processing algorithm for final ETT tip and Carina localization.
• Evaluation using three performance metrics on a clinical dataset.

Main Results:

• The proposed method achieved an 88.82% accuracy in detecting ETT malposition.
• The system demonstrated minimal ETT-Carina distance errors, averaging 5.333±6.240 mm.
• The automated approach successfully addressed limitations of previous methods, including manual intervention requirements.

Conclusions:

• The developed automated system offers a highly accurate and reliable solution for detecting endotracheal tube malposition.
• This AI-driven approach has the potential to significantly enhance patient safety in ICUs by ensuring correct ETT placement.
• The method's ability to function without manual intervention makes it a promising tool for routine clinical use.