Related Experiment Video
Updated: Dec 6, 2025
05:33
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
608
Synthesis of Prostate MR Images for Classification Using Capsule Network-Based GAN Model.
Houqiang Yu1,2, Xuming Zhang1
1Ministry of Education Key Laboratory of Molecular Biophysics, Department of Biomedical Engineering, School of Life Science and Technology, Huazhong University of Science and Technology, No 1037, Luoyu Road, Wuhan 430074, China.
Sensors (Basel, Switzerland)
|October 14, 2020
Summary
This study introduces a novel Generative Adversarial Network (GAN) using capsule networks to create high-quality synthetic prostate MR images, addressing data scarcity for deep learning diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Prostate cancer diagnosis relies heavily on MR imaging, but a lack of annotated data hinders deep learning model training.
- Existing data augmentation methods, including traditional and Generative Adversarial Network (GAN)-based approaches, often fail to generate diverse and high-quality synthetic MR images.
Purpose of the Study:
- To develop a novel Generative Adversarial Network (GAN) model for synthesizing realistic and diverse prostate MR images.
- To improve the quality and quantity of training data for deep learning-based prostate cancer diagnosis.
Main Methods:
- A novel GAN model was proposed, integrating capsule networks into the discriminator to enhance equivariant representations and robustness to spatial variations.
- The model utilizes deep convolutional GAN architecture with least squares loss to mitigate the vanishing gradient problem.
- Experiments were conducted using both simulated and real prostate MR images.
Main Results:
- The proposed capsule network-based GAN model generated MR images of superior realism and quality compared to existing GANs.
- Quantitative evaluations showed the proposed GAN achieved minimal Kullback-Leibler divergence for image generation.
- Incorporating the generated images into a deep learning classification task resulted in the best classification performance among evaluated models.
Conclusions:
- The novel capsule network-based GAN effectively synthesizes high-quality, diverse MR images, addressing the bottleneck of data scarcity in prostate cancer diagnosis.
- This approach enhances the performance of deep learning models for prostate cancer classification, offering a promising solution for improved diagnostic accuracy.
