الفهرس 
Chapter 1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 114 |  |  |
| | | from | 114 | | |
Abstract
Diagnosis of fundus diseases and commencing treatment without delay are key factors in averting irreversible vision loss in patients. Fundus imaging, a primary technique for retinal imaging, is recognized for its effectiveness in capturing and investigating anatomical features and abnormalities in the human eye. It plays a pivotal role in observing and identifying various ophthalmological diseases, which often manifest as changes in or around structures like the optic disk and blood vessels. Given the intricate nature of these conditions, it’s common for a patient’s fundus images to reveal single or multiple diseases affecting one or both eyes.
This study aims to explore and introduce multiple strategies for the detection and classification of ophthalmological diseases using different deep learning vision transformer approaches. We propose two strategies that deepen the understanding of these transformer approaches in disease detection. The first strategy leverages the robustness of pre-trained Swin Transformer V2, an advanced vision transformer method, while the second strategy employs the Data Efficient Image Transformer (DeiT), another model of vision transformers.
All models were trained and evaluated using the ODIR-2019 dataset, which contains fundus images from both left and right eyes. This dataset comprises eight categories, and the goal was to apply the transformer approach for multi-label classification. These images underwent different preprocessing and fine-tuning techniques to optimize the performance of the deep learning models.
The results of our study demonstrate that the proposed transformer approaches outperform state-of-the-art Convolutional Neural Network (CNN) models, achieving an accuracy of 93.1% and an Area Under the Curve (AUC) of 94.4% for multi-label ophthalmological disease classification. This shows that our methods not only improve performance but also provide a more comprehensive approach to diagnosing multiple eye conditions simultaneously.