Related Experiment Video
Updated: Jun 16, 2026
10:22
Fabrication and Characterization of Optical Tissue Phantoms Containing Macrostructure
Published on: February 12, 2018
10.8K
In-silico CT simulations of deep learning generated heterogeneous phantoms
Cornelio Salvador Salinas1, Kirti Magudia2, Aman Sangal3
1Center for Virtual Imaging Trials, Carl E. Ravin Advanced Imaging Laboratories, Duke University, United States of America.
Biomedical Physics & Engineering Express
|June 30, 2025
Summary
This study introduces a deep learning model to create realistic virtual imaging phantoms with intra-organ textures. The generated heterogeneous phantoms improve the fidelity of in silico trials for medical imaging simulations.
Area of Science:
- Medical Imaging
- Computational Phantoms
- Artificial Intelligence
Background:
- Current virtual imaging phantoms lack realistic intra-organ texture and material variation.
- Biological tissues exhibit inherent heterogeneity, necessitating more complex virtual models.
Purpose of the Study:
- To develop and train a deep learning model for generating realistic heterogeneous virtual imaging phantoms.
- To incorporate intra-organ texture and material variation into virtual phantoms for enhanced realism.
Main Methods:
- Training two 3D Double U-Net conditional generative adversarial networks (3D DUC-GAN) on CT image-segmentation pairs.
- Generating sixteen unique organ textures for torso organs.
- Simulating virtual CT scans of generated phantoms using DukeSim.
Main Results:
- The 3D DUC-GAN model synthesized realistic heterogeneous phantoms with a mean absolute difference of 46.15 ± 1.06 HU compared to original CT scans.
- Achieved a structural similarity index (SSIM) of 0.86 ± 0.004 and peak signal-to-noise ratio (PSNR) of 28.62 ± 0.14.
- Demonstrated significant improvements over homogeneous texture methods, with metrics showing 27-28% enhancement.
Conclusions:
- The generated heterogeneous phantoms offer a significant step toward more realistic in silico trials.
- Enhanced simulation of imaging procedures with greater fidelity to true anatomical variation is now possible.
- The deep learning approach provides a robust method for creating anatomically accurate and texturally realistic virtual phantoms.
