Related Experiment Video
Updated: Aug 21, 2025
09:32
Subject-specific Musculoskeletal Model for Studying Bone Strain During Dynamic Motion
Published on: April 11, 2018
9.8K
Simulation of Random Deformable Motion in Soft-Tissue Cone-Beam CT with Learned Models
Y Hu1, H Huang2, J H Siewerdsen2
1Dept. of Computer Science, Johns Hopkins University, Baltimore, MD, USA.
Proceedings of Spie--The International Society for Optical Engineering
|November 16, 2022
Summary
This study introduces a new framework for simulating realistic motion in Cone-beam CT (CBCT) scans, crucial for improving interventional radiology. The method uses generative adversarial networks (GANs) to create complex motion patterns for better training of motion compensation techniques.
Area of Science:
- Medical Imaging
- Radiology
- Machine Learning
Background:
- Cone-beam CT (CBCT) is vital for interventional radiology guidance but suffers from motion artifacts.
- Existing motion compensation methods struggle with complex, combined motion patterns like respiratory and peristaltic movements.
- Simulating realistic, complex motion for training deep learning models remains a significant challenge.
Purpose of the Study:
- To develop a framework for synthesizing realistic deformable motion trajectories in soft-tissue CBCT.
- To enable unsupervised training of deep learning models using unpaired clinical CBCT data.
- To validate the feasibility of generating variable CBCT data with realistic motion for improved motion compensation.
Main Methods:
- Utilized conditional generative adversarial networks (GANs) to learn complex motion from unlabeled CBCT volumes.
- Developed a framework for unsupervised learning with unpaired clinical CBCT data.
- Conducted a feasibility study using simulated data with known motion for controlled validation.
Main Results:
- The proposed framework successfully generated realistic and variable CBCT deformable motion fields.
- Synthetic motion induced diffeomorphic deformations (Jacobian Determinant > 0).
- Demonstrated accurate displacement patterns (0.5 mm in static regions, 3.8 mm in dynamic regions) and directional bias (superior-inferior motion).
Conclusions:
- The framework shows feasibility for realistic motion simulation in CBCT.
- Enables the synthesis of variable CBCT data for training advanced motion compensation algorithms.
- Paves the way for improved accuracy in image-guided interventions by addressing motion artifacts.
