Fundus GAN - GAN-based Fundus Image Synthesis for Training Retinal Image Classifiers.
Summary
Fundus GAN generates synthetic retinal images, overcoming data limitations and patient privacy concerns for AI-driven eye disease diagnosis. This method enhances the training of diagnostic classifiers without compromising patient data.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Deep learning for fundus image analysis faces challenges due to limited labeled data and patient privacy concerns.
- Existing data augmentation methods are insufficient and do not address privacy issues.
- Synthetic data generation is crucial for developing robust AI diagnostic tools.
Purpose of the Study:
- To propose a novel Generative Adversarial Network-based (GAN-based) method, Fundus GAN, for synthesizing realistic fundus images.
- To address the scarcity of privacy-preserving training data for AI-based eye disease diagnosis.
- To improve the performance of retinal image classifiers.
Main Methods:
- A two-step GAN-based approach involving vessel tree extraction via a segmentation network.
- Unsupervised generative attention networks for vessel tree to fundus image-to-image translation.
- Utilizing synthetic images to train retinal image classifiers.
Main Results:
- Fundus GAN demonstrates superior performance compared to state-of-the-art methods across various evaluation metrics.
- Generated retinal images effectively trained classifiers for eye disease diagnosis.
- The method successfully generates privacy-preserving synthetic training data.
Conclusions:
- Fundus GAN offers a viable solution to the limited and privacy-sensitive data challenge in AI-driven fundus image analysis.
- The proposed synthesis method can significantly enhance the accuracy of eye disease diagnostic classifiers.
- This approach facilitates the development of automated diagnostic systems while upholding patient privacy.
