An automatic glaucoma grading method based on attention mechanism and EfficientNet-B3 network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible blindness, affecting millions globally, with an expected rise to over 110 million patients by 2040.
• Early glaucoma detection is critical to prevent vision loss, yet current methods often fail to distinguish between disease stages.
• Accurate glaucoma grading is essential for timely and appropriate clinical intervention.

Purpose of the Study:

• To develop an automated deep learning method for precise glaucoma classification into normal, early, and moderately advanced stages.
• To enhance feature extraction and classification accuracy using multimodal data (2D fundus and 3D-OCT images).
• To improve the efficiency and reliability of glaucoma diagnosis through advanced AI techniques.

Main Methods:

• A novel automatic glaucoma classification method utilizing the EfficientNetB3 network and an attention mechanism.
• Fusion of features extracted from 2D fundus images (ResNet34) and 3D-OCT scanner images (EfficientNetB3).
• Implementation of a two-branch model with attention to focus on salient ocular features and minimize redundancy.

Main Results:

• The proposed method achieved a highest accuracy of 97.83% in classifying glaucoma stages.
• The attention mechanism effectively focused on relevant image features, improving model performance.
• The model demonstrated minimized feature redundancy and enhanced classification accuracy.

Conclusions:

• The developed automatic grading method is effective for reliable glaucoma screening.
• This AI tool can serve as a valuable second opinion for clinicians, reducing diagnostic errors.
• Improved glaucoma classification can significantly aid in timely patient treatment and prevent vision loss.