Related Experiment Video
Updated: Dec 3, 2025
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
9.3K
Automatic segmentation of prostate magnetic resonance imaging using generative adversarial networks.
Wei Wang1, Gangmin Wang2, Xiaofen Wu3
1Department of Radiology, Tongji Hospital of Tongji University School of Medicine, Shanghai, China.
Clinical Imaging
|October 29, 2020
Summary
We developed SegDGAN, an automatic prostate segmentation method using generative adversarial networks. This AI model significantly improves the accuracy of prostate gland segmentation in MRI scans, outperforming existing methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Prostate segmentation from MRI is crucial for diagnosis but traditionally manual and time-consuming.
- Manual segmentation requires significant clinical expertise and observer experience.
- Automatic segmentation methods aim to improve efficiency and consistency.
Purpose of the Study:
- To introduce SegDGAN, an automated prostate segmentation method based on generative adversarial networks (GANs).
- To evaluate the accuracy and performance of SegDGAN compared to existing segmentation techniques.
- To provide an open-source implementation for wider research adoption.
Main Methods:
- Proposed SegDGAN utilizes a fully convolutional generation network with densely connected blocks and a critic network with multi-scale feature extraction.
- Objective function optimized using mean absolute error and Dice coefficient for enhanced segmentation accuracy.
- Compared SegDGAN against U-Net, FCN, and SegAN on a 220-patient dataset and public datasets using DSC, VOE, ASD, and HD metrics.
Main Results:
- SegDGAN achieved the highest Dice Similarity Coefficient (DSC) of 91.66% on the clinical dataset.
- Achieved lowest Volume Overlap Error (VOE) of 15.28%, Average Surface Distance (ASD) of 0.51 mm, and Hausdorff Distance (HD) of 11.58 mm.
- On the PROMISE12 public dataset, SegDGAN obtained a DSC of 86.24% and the lowest VOE, ASD, and HD.
Conclusions:
- The SegDGAN model demonstrates significant potential for improving the accuracy of MRI-based prostate gland segmentation.
- Experimental results confirm SegDGAN's superior performance over traditional and other deep learning methods.
- The developed model offers a more efficient and accurate solution for prostate imaging analysis.
