Improved subcutaneous edema segmentation on abdominal CT using a generated adipose tissue density prior

Area of Science:

• Medical imaging analysis
• Computational pathology
• Biomedical engineering

Background:

• Edema, or swelling, is a common clinical sign associated with kidney, heart, and liver diseases.
• Accurate volumetric measurement of edema is clinically valuable for diagnosis and management.
• Subcutaneous adipose tissue is a frequent site for edema accumulation.

Purpose of the Study:

• To improve the accuracy of edema segmentation and volume measurement in subcutaneous adipose tissue.
• To develop a novel method for distinguishing edema from normal adipose tissue.
• To enhance the clinical utility of volumetric edema assessment.

Main Methods:

• Utilized a conditional generative adversarial network to create an adipose tissue mask excluding edema.
• Integrated the generated mask's density distribution into a Chan-Vese level set framework.
• Iteratively updated density distributions to separate edema and subcutaneous adipose tissue.

Main Results:

• Demonstrated significant improvement in edema segmentation accuracy on 25 patient datasets.
• Achieved an increase in the average Dice Similarity Coefficient from 56.0% to 61.7%.
• Reduced the average relative volume difference from 36.5% to 30.2% compared to previous methods.

Conclusions:

• The integration of a generated adipose tissue density prior significantly enhances edema segmentation.
• The improved segmentation accuracy supports the clinical utility of precise volumetric edema measurement.
• This approach offers a more accurate method for quantifying edema in subcutaneous adipose tissue.