کنفرانس بین المللی فناوری اطلاعات و دانش

صفحه اصلی / شانزدهمین کنفرانس بین المللی فناوری اطلاعات و دانش

A hybrid CNN–transformer framework for retinal disease classification

نویسندگان :

Hanie Zomorrodi¹ Hassan Khotanlou²

1- دانشگاه بوعلی سینا 2- دانشگاه بوعلی سینا

کلمات کلیدی :

convolutional neural network،retina،transformer encoder

چکیده :

Accurate diagnosis of retinal diseases is essential for preventing visual impairment and blindness. In this study, we propose a deep learning-based framework for automatic multi-class classification of retinal images that can detect 20 ocular diseases at once. The approach starts with preprocessing and improving fundus images, followed by data augmentation to boost the model’s generalization and strength. We extract features using a combined EfficientNet-ConvNeXt framework, which captures both local details and global context. Next, we refine the extracted features with a Transformer Encoder to model relationships across the entire retinal image. Finally, an MLP classifies the input. Experimental results show that our method achieves a Model Score of 0.903, surpassing earlier methods. These findings confirm that combining feature representations from the EfficientNet-ConvNeXt architecture with Transformer-based modeling significantly enhances the accuracy of retinal disease classification.