PHOTODIAGNOSIS WITH DEEP LEARNING: A GAN AND AUTOENCODER-BASED APPROACH FOR DIABETIC RETINOPATHY DETECTION

Photodiagnosis and Photodynamic Therapy, Journal Year: 2025, Volume and Issue: unknown, P. 104552 - 104552

Published: March 1, 2025

Diabetic retinopathy (DR) is a leading cause of visual impairment and blindness worldwide, necessitating early detection accurate diagnosis. This study proposes novel framework integrating Generative Adversarial Networks (GANs) for data augmentation, denoising autoencoders noise reduction, transfer learning with EfficientNetB0 to enhance the performance DR classification models. GANs were employed generate high-quality synthetic retinal images, effectively addressing class imbalance enriching training dataset. Denoising further improved image quality by reducing eliminating common artifacts such as speckle noise, motion blur, illumination inconsistencies, providing clean consistent inputs model. was fine-tuned on augmented denoised The achieved exceptional metrics, including 99.00% accuracy, recall, specificity, surpassing state-of-the-art methods. custom-curated OCT dataset featuring high-resolution clinically relevant challenges limited annotated noisy inputs. Unlike existing studies, our work uniquely integrates GANs, autoencoders, EfficientNetB0, demonstrating robustness, scalability, clinical potential proposed framework. Future directions include interpretability tools adoption exploring additional imaging modalities improve generalizability. highlights transformative deep in critical diabetic

Language: Английский

Data Augmentation with Generative Methods for Inherited Retinal Diseases: A Systematic Review

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(6), P. 3084 - 3084

Published: March 12, 2025

Inherited retinal diseases (IRDs) are rare and genetically diverse disorders that cause progressive vision loss affect 1 in 3000 individuals worldwide. Their rarity genetic variability pose a challenge for deep learning models due to the limited amount of data. Generative offer promising solution by creating synthetic data improve training datasets. This study carried out systematic literature review investigate use generative augment IRDs assess their impact on performance classifiers these diseases. Following PRISMA 2020 guidelines, searches four databases identified 32 relevant studies, 2 focused IRD rest other The results indicate effectively small Among techniques identified, Deep Convolutional Adversarial Networks (DCGAN) Style-Based Generator Architecture (StyleGAN2) were most widely used. These architectures generated highly realistic data, often indistinguishable from real even experts. highlight need more research into generation develop robust diagnostic tools studies comprehensive repositories.

Language: Английский