Related Experiment Video
Updated: Jul 21, 2026
10:59
Reconstruction of 3-Dimensional Histology Volume and its Application to Study Mouse Mammary Glands
Published on: July 26, 2014
14.5K
3D-MedTranCSGAN: 3D Medical Image Transformation using CSGAN.
1Department of Computing Technologies, School of Computing, SRM Institute of Science and Technology, Kattankulathur, India.
Computers in Biology and Medicine
|January 18, 2023
Summary
This study introduces a 3D medical image transformation model, 3D-MedTranCSGAN, which integrates Cyclic Synthesized Generative Adversarial Networks with non-adversarial losses. The model effectively performs various medical image transformations, outperforming existing methods in experimental evaluations.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Medical image transformation is crucial for various applications.
- Existing computer vision techniques face limitations in accuracy and scope.
- Generative Adversarial Networks (GANs) show promise but require refinement for medical imaging tasks.
Purpose of the Study:
- To propose a novel 3D medical image transformation model, 3D-MedTranCSGAN.
- To integrate non-adversarial loss components with Cyclic Synthesized Generative Adversarial Networks (CSGAN).
- To demonstrate the model's versatility across multiple medical image transformation tasks.
Main Methods:
- Developed the 3D-MedTranCSGAN model, incorporating a PatchGAN discriminator.
- Utilized non-adversarial loss functions including content, perception, and style transfer losses.
- Introduced a 3DCascadeNet generator architecture for enhanced image perceptiveness.
Main Results:
- Achieved superior performance in PET to CT transformation, CT to PET reconstruction, MR motion artifact modification, and PET noise removal.
- Demonstrated high quantitative metrics across tasks, including SSIM, PSNR, MSE, VIF, UQI, and LPIPs.
- Outperformed existing transformation methods in all experimental evaluations.
Conclusions:
- The 3D-MedTranCSGAN model offers a robust and versatile solution for diverse 3D medical image transformation challenges.
- The integration of CSGAN with specific loss functions and a novel generator architecture significantly enhances transformation quality.
- The model's effectiveness across multiple tasks highlights its potential for widespread adoption in medical imaging.
