Robust Body Composition Analysis by Generating 3D CT Volumes from Limited 2D Slices

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Body composition analysis is crucial for understanding aging and disease.
• Traditional 2D CT scans for analysis have spatial variability issues.
• Radiation exposure concerns limit repeated 2D CT imaging.

Purpose of the Study:

• To develop a novel method for generating 3D CT volumes from limited 2D slices.
• To improve the accuracy and robustness of body composition analysis.
• To mitigate spatial variability inherent in 2D CT imaging.

Main Methods:

• Utilized a latent diffusion model (LDM) to generate 3D CT volumes.
• Employed a variational autoencoder to map 2D slices into a latent space.
• Incorporated body part regression for accurate slice interpolation and 3D volume construction.

Main Results:

• The proposed method significantly enhances body composition analysis.
• Reduced the error rate in body composition analysis from 23.3% to 15.2%.
• Demonstrated effectiveness on both in-house and public 3D abdominal CT datasets.

Conclusions:

• The LDM-based approach offers a more accurate alternative to traditional 2D CT analysis.
• This method can improve insights into aging and disease progression.
• It addresses limitations of spatial variability and radiation exposure in CT-based analysis.