Anatomically Guided Deep Learning System for Right Internal Jugular Line (RIJL) Segmentation and Tip Localization in...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Radiology

Background:

• Right internal jugular lines (RIJL) are crucial central venous catheters (CVCs) in ICUs.
• Accurate RIJL placement verification via chest X-rays (CXRs) is vital for patient safety.
• Automated detection of CVCs in CXRs is developing, but RIJL-specific segmentation and tip localization require further research.

Purpose of the Study:

• To develop and evaluate a deep learning system for accurate segmentation of the RIJL course and precise localization of its tip in CXR images.
• To improve upon existing methods for RIJL analysis, addressing the paucity of focused investigations.

Main Methods:

• A deep learning system was developed using the nnU-Net framework, integrating anatomical landmark segmentation (trachea) and a dedicated RIJL segmentation network.
• The system utilized subregion extraction based on the trachea landmark to focus the segmentation network on relevant image areas.
• Customized postprocessing was applied to refine segmentation results and define the RIJL tip based on the most inferior point.

Main Results:

• The proposed deep learning system significantly improved RIJL segmentation and tip localization compared to a baseline network.
• Mean Average Symmetric Surface Distance (ASSD) for segmentation decreased from 2.72 mm to 1.41 mm.
• Mean tip distance error was reduced from 11.27 mm to 8.29 mm.

Conclusions:

• The integrated deep learning system, leveraging anatomical landmarks and custom postprocessing, demonstrates enhanced accuracy for RIJL segmentation and tip localization in CXRs.
• This AI-driven approach shows promise for improving the efficiency and precision of RIJL verification in clinical settings.
• The findings suggest a valuable tool for reducing clinician workload and ensuring optimal patient care through accurate CVC monitoring.