Enhancing Glaucoma Diagnosis Through Multi-Layer Transformer and Multi-Modal Feature Fusion.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis often relies on single examination modalities, limiting comprehensive assessment.
• Existing systems struggle with the complexity and nuances of glaucoma progression.

Purpose of the Study:

• To develop an advanced glaucoma grading framework integrating multiple examination modalities.
• To overcome limitations of single-modality diagnostic systems for enhanced glaucoma diagnosis.

Main Methods:

• Proposed a novel multi-modal glaucoma grading framework (M2F3) simulating clinical diagnosis.
• Utilized a multi-layer transformer (MLT) for efficient fusion of multi-modal data.
• Incorporated contrastive learning and prior ocular structure knowledge for improved feature learning.

Main Results:

• The M2F3 framework achieved a 0.0465 increase in Cohen's kappa (κ) coefficient.
• Demonstrated superior performance compared to state-of-the-art (SOTA) methods on the GAMMA dataset.

Conclusions:

• The multi-modal framework provides a more accurate glaucoma diagnostic tool.
• Integration of multiple modalities and prior knowledge significantly improves upon single-modality systems.