Ultrasound Probe Pose Classification for Task Recognition in Central Venous Catheterization.

Area of Science:

• Medical Simulation
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Central Line Tutor enhances central venous catheterization training with real-time feedback.
• Current systems rely on costly electromagnetic tracking for ultrasound image analysis.
• This tracking hinders accessibility and increases training costs.

Purpose of the Study:

• To validate deep learning models for vessel segmentation and ultrasound probe pose classification.
• To reduce reliance on electromagnetic tracking in medical simulation systems.
• To improve the accessibility and affordability of advanced medical training tools.

Main Methods:

• Developed a U-Net architecture for neck vasculature segmentation from ultrasound images.
• Designed a Convolutional Neural Network (CNN) for ultrasound probe pose classification.
• Evaluated a combined U-Net and CNN approach for enhanced classification accuracy.

Main Results:

• U-Net achieved a mean Intersect over Union score of 0.746 ± 0.052 for vessel segmentation.
• CNN achieved a mean classification accuracy of 92.0% ± 3.0 for probe pose.
• The combined U-Net + CNN model reached 92.7% ± 2.1% accuracy.

Conclusions:

• Deep learning methods show significant potential to replace electromagnetic tracking for ultrasound analysis in medical training.
• Validated AI models can accurately segment vessels and classify ultrasound probe pose.
• This research paves the way for more cost-effective and accessible central venous catheterization training simulators.