GLEAM: A Multimodal Imaging Dataset and HAMM for Glaucoma Classification

A consortium of researchers has published a breakthrough in medical AI with the release of GLEAM (Glaucoma Lesion Evaluation and Analysis with Multimodal Imaging), the first publicly available dataset combining three critical eye imaging modalities for glaucoma diagnosis. The dataset includes scanning laser ophthalmoscopy fundus images, circumpapillary OCT (optical coherence tomography) images, and visual field pattern deviation maps, all annotated across four progressive disease stages. This tri-modal approach allows AI systems to leverage complementary information that single-image systems miss, providing a more complete picture of optic nerve damage and visual field loss.

To effectively process this complex data, the team developed HAMM (Hierarchical Attentive Masked Modeling), a novel AI framework specifically designed for multimodal medical classification. HAMM employs hierarchical attentive encoders with lightweight decoders, focusing the representation learning on the encoder side to better integrate cross-modal signals. This architecture allows the model to learn which imaging modality provides the most relevant information at different disease stages, potentially enabling earlier detection and more precise staging than current clinical methods. The entire package—dataset and model—is now available to the research community via arXiv, accelerating development of diagnostic tools for a disease affecting over 80 million people worldwide.