Glaucoma Diagnosis and Progression Prediction Based on Deep Multimodal Fusion and Collaborative Attention Mechanism

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a primary cause of irreversible blindness globally.
• Current diagnostic methods often use single data types, missing crucial multimodal insights.
• Poor image quality in clinical fundus photography presents a significant challenge.

Purpose of the Study:

• To develop an advanced model for glaucoma diagnosis and progression prediction.
• To address limitations of unimodal data and poor image quality in current approaches.
• To effectively fuse diverse data modalities for enhanced diagnostic performance.

Main Methods:

• Proposed MUGLDEM model utilizing deep multimodal fusion and collaborative attention.
• Integrated clinical text, fundus photography, and OCT data from the UK Biobank.
• Employed image enhancement techniques (denoising, histogram equalization) for low-quality fundus images.
• Developed independent feature extraction for each modality.
• Implemented dual-gate residual and bidirectional cross-attention for feature fusion.

Main Results:

• MUGLDEM achieved state-of-the-art performance on glaucoma diagnosis tasks.
• The model demonstrated superior accuracy in predicting glaucoma progression.
• Experimental validation was conducted using the UK Biobank dataset.

Conclusions:

• Deep multimodal fusion with collaborative attention offers a powerful approach for glaucoma management.
• MUGLDEM effectively overcomes challenges related to data quality and integration.
• The model shows significant promise for improving patient outcomes in glaucoma care.