Enhancing glaucoma diagnosis: Generative adversarial networks in synthesized imagery and classification with...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible vision loss, often asymptomatic in early stages.
• Accurate and early diagnosis is crucial but challenging with conventional methods.
• Medical imaging datasets for glaucoma often suffer from imbalance and small sample sizes.

Purpose of the Study:

• To develop and validate a novel deep learning approach for accurate glaucoma diagnosis.
• To address data scarcity and quality issues in medical imaging datasets using Generative Adversarial Networks (GANs).
• To improve the classification accuracy of glaucoma detection using a pretrained MobileNetV2 model.

Main Methods:

• Generative Adversarial Networks (GANs) were employed to synthesize high-quality, realistic fundus images, enhancing dataset diversity.
• Gaussian Filtering was applied for noise reduction in fundus images.
• An enhanced level set algorithm was utilized for precise optic cup segmentation.
• A pretrained MobileNetV2 architecture was fine-tuned for binary classification of normal versus glaucoma images.

Main Results:

• The proposed method achieved a high classification accuracy of 98.9% in differentiating normal and glaucoma images.
• GANs effectively generated high-quality fundus images, mitigating dataset imbalances.
• The enhanced level set algorithm provided accurate segmentation of glaucomatous features.

Conclusions:

• The developed approach demonstrates the significant potential of GANs and transfer learning in medical imaging for glaucoma screening.
• This AI-driven method offers a promising tool for early and accurate glaucoma diagnosis, potentially reducing vision loss.
• The study highlights the effectiveness of combining advanced image generation, segmentation, and deep learning classification techniques.