Related Experiment Video
Updated: Jul 5, 2025
08:41
Patient-Specific Polyvinyl Alcohol Phantom Fabrication with Ultrasound and X-Ray Contrast for Brain Tumor Surgery Planning
Published on: July 14, 2020
8.5K
Medical image synthesis via conditional GANs: Application to segmenting brain tumours
Mohammad Hamghalam1, Amber L Simpson2
1School of Computing, Queen's University, Kingston, ON, Canada; Department of Electrical Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran.
Computers in Biology and Medicine
|January 24, 2024
Summary
This study introduces novel generative adversarial networks (GANs) to improve brain tumour segmentation accuracy from MRI scans. The models enhance image contrast, leading to better identification of tumour subregions for diagnosis and surgical planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Accurate brain tumour segmentation is vital for clinical applications like surgical planning and diagnosis.
- Magnetic resonance imaging (MRI) is preferred for brain tissue visualization, but intensity variations hinder segmentation accuracy.
- Overlapping tissue densities in MRI scans reduce contrast, posing challenges for precise tumour delineation.
Purpose of the Study:
- To introduce a novel framework using conditional generative adversarial networks (cGANs) to enhance tumour subregion contrast.
- To improve voxel-wise and region-wise brain tumour segmentation accuracy.
- To develop models that address the challenge of overlapping intensity distributions in MRI scans.
Main Methods:
- Developed two models: Enhancement and Segmentation GAN (ESGAN) and Enhancement GAN (EnhGAN).
- ESGAN combines classifier and adversarial loss for patch-based label prediction.
- EnhGAN generates high-contrast synthetic images using a novel adaptive voxel calibration generator and a multi-scale Markovian discriminator.
Main Results:
- The proposed models demonstrate competitive accuracy compared to existing brain tumour segmentation techniques.
- ESGAN and EnhGAN effectively enhance contrast and reduce inter-class overlap in MR images.
- Experimental results on public datasets validate the efficacy of the developed framework.
Conclusions:
- The novel cGAN-based framework significantly improves brain tumour segmentation accuracy.
- The developed models offer a promising approach for enhancing contrast in medical imaging for better diagnostic outcomes.
- This work contributes to advancing automated analysis of brain tumours using deep learning techniques.
