GLEAM: A Multimodal Imaging Dataset and HAMM for Glaucoma Classification.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a primary cause of irreversible blindness globally.
• Early glaucoma stages are often asymptomatic, delaying diagnosis and treatment.
• Current diagnostic methods often rely on single or dual imaging modalities, limiting information integration.

Purpose of the Study:

• To introduce the first publicly available tri-modal glaucoma dataset (GLEAM) for enhanced diagnosis.
• To develop a novel multimodal deep learning framework (HAMM) for glaucoma classification.
• To leverage complementary information from fundus, OCT, and visual field data for accurate, stage-specific glaucoma detection.

Main Methods:

• Development of the Glaucoma Lesion Evaluation and Analysis with Multimodal Imaging (GLEAM) dataset, including fundus, OCT, and visual field data.
• Proposal of Hierarchical Attentive Masked Modeling (HAMM) for multimodal glaucoma classification.
• Implementation of Multimodal-Channel Graph Attention (MCGA) to weigh and integrate cross-modal information based on clinical reasoning.

Main Results:

• Tri-modal data fusion significantly outperforms single-modal and dual-modal approaches on the GLEAM dataset.
• The proposed HAMM framework achieves superior performance compared to existing state-of-the-art multimodal learning methods.
• The MCGA module effectively emulates clinical reasoning for improved classification accuracy.

Conclusions:

• The GLEAM dataset and HAMM framework facilitate more accurate and earlier diagnosis of glaucoma.
• Integrating multimodal imaging data is crucial for overcoming limitations of single-modality approaches.
• The developed methods enable effective stage-specific treatment strategies for glaucoma patients.