Diffusion Model-Based Displacement Field Generation for 4D-CT Chest Image Generation.

Area of Science:

• Medical Imaging
• Radiotherapy
• Computational Anatomy

Background:

• Accurate individual-level respiratory motion data is crucial for enhancing surgical and radiotherapy precision.
• Current time-series imaging methods like 4D CT and deep learning interpolation pose burdens due to breath-holding and radiation exposure.

Purpose of the Study:

• To develop a framework for generating four-dimensional (4D) computed tomography (CT) images from a single-phase 3D CT scan.
• To utilize a conditional diffusion model for generating displacement vector fields (DVFs) representing respiratory motion.
• To enable 4D CT generation using only the magnitude of displacement, reducing patient invasiveness.

Main Methods:

• A conditional diffusion model was employed to generate displacement vector fields (DVFs).
• The model incorporated the initial-phase CT image and the mean DVF of the target phase as guidance.
• The framework was trained and tested on 4D-CT images from 62 cases.

Main Results:

• The proposed model successfully generated 4D CT images from single-phase 3D CT scans.
• Quantitative comparisons confirmed the validity of the approach under various guidance scenarios.
• Predicted DVFs accurately captured respiratory motion, enabling effective deformation of CT images.

Conclusions:

• The diffusion model effectively predicts DVFs for respiratory motion, facilitating 4D CT generation from single scans.
• This approach can be directly applied to radiotherapy planning.
• The method is expected to improve radiation targeting accuracy by providing detailed respiratory motion data.