A Hybrid Approach to Glaucoma Disease Prediction Using Vision Transformer Model

Abstract

Glaucoma, a leading cause of permanent vision loss globally, can be effectively managed with early detection, making timely diagnosis crucial for preserving sight. The paper proposed a hybrid model combining Bidirectional Long Short-Term Memory (BiLSTM) and Enhanced Vision Transformer (EViT) for automated glaucoma detection in fundus images. The BiLSTM captures temporal dependencies, while the EViT leverages spatial relationships, improving performance. The specific methodology consists of the following steps: (1) Image Acquisition; (2) Image preprocessing with data augmentation; (3) Hybrid BiLSTM with Enhanced Vision Transformer Learning for Glaucoma Disease Prediction; (4) experimental evaluations and comparisons with conventional deep learning models to validate the efficacy and utility of the proposed hybrid model for Glaucoma prediction. The proposed method achieves state-of-the-art performance on the RIM-ONE DL image dataset, with impressive metrics: 97% precision, 96.7% recall, 97.8% accuracy, and 96.62% F1-score, surpassing existing CNN-based and attention-based glaucoma detection approaches.