Related Experiment Video
Updated: Aug 7, 2026
Gene Regulation and Targeted Therapy in Gastric Cancer Peritoneal Metastasis: Radiological Findings from Dual Energy CT and PET/CT
Published on: January 22, 2018
Radiomics using generative adversarial network enhanced non-contrast computed tomography for gastric cancer diagnosis
Xiaodong Li1, Yunpeng Zhao1, Mengjie Fang2
1School of Artificial Intelligence, University of the Chinese Academy of Sciences, No. 19 (A), Yuquan Road, Shijingshan District, Beijing, China, Beijing, Beijing, 100049, China.
This study introduces a new framework using Generative Adversarial Networks (GANs) to create synthetic contrast-enhanced CT (CECT) images from non-contrast-enhanced CT (NCCT) scans. This improves gastric cancer diagnosis accuracy without the risks of traditional CECT scans.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Non-contrast-enhanced CT (NCCT) has limitations in tissue resolution for gastric cancer diagnosis.
- Contrast-enhanced CT (CECT) poses risks like allergic reactions and high radiation exposure.
- There is a need for safer, more effective imaging methods for gastric cancer evaluation.
Purpose of the Study:
- To develop an auxiliary diagnostic framework integrating Generative Adversarial Networks (GANs) with radiomics.
- To enhance the diagnostic utility of NCCT images for gastric cancer.
- To enable informative imaging-based evaluations without contrast agents.
Main Methods:
- Collected NCCT, CECT, and clinical data from 1,757 gastric cancer patients across four centers.
- Developed and optimized Pix2Pix-based GAN configurations with attention mechanisms and a composite loss function.
- Applied Symmetric Normalization-based registration for image alignment and quantitative metrics (MAE, MSE, PSNR, SSIM) for model evaluation.
Main Results:
- The proposed GAN model significantly outperformed existing methods (Pix2Pix, CycleGAN, BBDM) on internal and external test sets.
- Synthetic CECT (SCECT) images generated by the model improved radiomics-based prediction of histological grade, Lauren classification, and T stage.
- Area Under the Curve (AUC) improvements ranged from 0.7% to 7.3% for various diagnostic parameters compared to NCCT alone.
Conclusions:
- The proposed framework effectively enhances the diagnostic value of NCCT images for gastric cancer.
- It provides a safer and more efficient auxiliary tool by eliminating the need for contrast agents.
- This approach facilitates more informative imaging-based gastric cancer evaluations.
