Beyond the lungs: extending the field-of-view in chest CT with latent diffusion models

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Chest CT imaging often has a restricted field of view (FOV), limiting analysis of lung disease impact on other organs.
• Understanding inter-organ relationships is vital for assessing lung disease risks and patient outcomes.
• Current imaging practices prioritize lung-specific views due to cost and radiation concerns.

Purpose of the Study:

• To develop a novel approach, SCOPE (Spatial Coverage Optimization with Prior Encoding), to extend the FOV of chest CT images.
• To capture inter-organ relationships by generating new axial slices beyond the original FOV.
• To enable comprehensive analysis of how lung diseases affect organs like the liver and kidneys.

Main Methods:

• Trained a variational autoencoder (VAE) to encode 2D axial CT slices.
• Stacked VAE latent representations to create a 3D context for a latent diffusion model.
• Utilized the trained model for zero-shot generation of new axial slices to extend the FOV in the z-direction.

Main Results:

• SCOPE effectively extended the FOV in chest CT images to include the liver and kidneys.
• Generated slices demonstrated high fidelity with acquired data, achieving a Structural Similarity Index (SSIM) of 0.81 on a whole-body dataset.
• The approach was validated on the National Lung Screening Trial (NLST) dataset.

Conclusions:

• SCOPE offers a promising solution for extending CT image FOV without additional scanning.
• This method can enhance the understanding of systemic effects of lung diseases.
• The developed technique facilitates more comprehensive patient care by revealing inter-organ impacts.