Artificial intelligence versus expert: a comparison of rapid visual inferior vena cava collapsibility assessment...

Area of Science:

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

Background:

• Inferior vena cava (IVC) collapsibility assessment via point-of-care ultrasound (POCUS) is crucial for fluid management in critically ill patients.
• Novice POCUS providers may face challenges in accurately determining IVC collapsibility.
• Developing automated tools can enhance diagnostic capabilities and support clinical decision-making.

Purpose of the Study:

• To develop and evaluate a deep learning algorithm for quantifying IVC collapsibility.
• To assess the algorithm's agreement with expert POCUS providers' assessments.
• To create a tool that assists novice POCUS users in real-time IVC analysis.

Main Methods:

• A long short-term memory (LSTM) deep learning architecture was employed for real-time ultrasound video analysis.
• The algorithm was trained on 220 public domain IVC ultrasound videos, with data augmentation techniques used.
• Performance was evaluated against 50 new IVC ultrasound videos, comparing algorithm-expert agreement using Fleiss' κ.

Main Results:

• Expert POCUS providers showed very substantial agreement in assessing IVC collapsibility (κ = 0.65).
• The deep learning algorithm achieved moderate agreement with expert assessments (κ = 0.45).
• The algorithm demonstrated potential for real-time IVC collapsibility determination.

Conclusions:

• The developed deep learning algorithm shows good agreement with POCUS experts for estimating IVC collapsibility.
• This AI tool can aid in differentiating fluid-responsive from fluid-unresponsive septic shock.
• The algorithm could be integrated into ultrasound machines to support novice POCUS providers, simplifying IVC assessment.