Related Experiment Video
Updated: May 10, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
Dual-Stream Contrastive Latent Learning Generative Adversarial Network for Brain Image Synthesis and Tumor
Junaid Zafar1, Vincent Koc2, Haroon Zafar3
1Faculty of Engineering, Government College University, Lahore 54000, Pakistan.
Journal of Imaging
|April 25, 2025
Summary
This study introduces a novel dual-stream generative adversarial network (DS-GAN) for enhanced MRI image augmentation. The DS-GAN effectively synthesizes diverse and high-fidelity medical images, improving AI-driven diagnostics.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Machine Learning
Background:
- Generative Adversarial Networks (GANs) often focus on pixel-level details, limiting their ability to capture overall image distribution crucial for medical image synthesis.
- Robust augmentation of Magnetic Resonance Imaging (MRI) data is essential for improving the performance of AI-driven diagnostic systems.
Purpose of the Study:
- To develop a novel generative adversarial network (GAN) architecture for robust augmentation of MRI images.
- To enhance the synthesis of diverse and high-fidelity medical images by addressing limitations in current GANs.
Main Methods:
- Proposed a dual-stream contrastive latent projection generative adversarial network (DSCLPGAN) with specialized pathways for local and global feature modeling.
- Integrated a transformer-based encoder-decoder framework and a contrastive learning projection (CLP) module for contextual coherence and latent space diversity.
- Employed an ensemble of specialized discriminators (D1, D2, D3) for classification consistency, localized variation mapping, and structural preservation.
Main Results:
- Achieved state-of-the-art performance on a dataset of 3064 T1-weighted contrast-enhanced MRI scans with brain tumors.
- Obtained a Structural Similarity Index (SSIM) of 0.99, a classification accuracy of 99.4%, and a Peak Signal-to-Noise Ratio (PSNR) of 34.6 dB.
- Demonstrated high-fidelity augmentation capabilities, significantly improving image synthesis quality.
Conclusions:
- The proposed DSCLPGAN effectively generates diverse and high-fidelity MRI augmentations, outperforming existing methods.
- This approach holds significant potential for enhancing AI-driven clinical decision support systems by providing reliable augmented datasets.
- The dual-stream architecture and contrastive learning contribute to comprehensive image synthesis and improved diagnostic accuracy.
