Fundus Image-based Glaucoma Screening via Retinal Knowledge-Oriented Dynamic Multi-Level Feature Integration

2026-04-14

AI Summary

This study addresses the limited cross-domain generalization of existing data-driven glaucoma screening models, which often lack explicit modeling of retinal anatomical structures and rely on fixed-region feature extraction that struggles with lesion heterogeneity. To overcome these limitations, the authors propose a dynamic multi-level feature integration framework that incorporates retinal prior knowledge. The approach employs a three-branch network to jointly model global context, optic disc/cup morphology, and dynamically localized pathological regions. Crucially, anatomical priors derived from a pretrained foundation model are innovatively embedded into an attention mechanism, enabling dynamic windowing that adaptively focuses on diagnostically critical areas. Evaluated on the AIROGS dataset, the method achieves 98.5% AUC and 94.6% accuracy, and demonstrates strong cross-domain generalization across multiple datasets in the SMDG-19 benchmark.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multi-instance/Multi-view Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Security and Privacy: Data transparency and provenance

Abstract

Automated diagnosis based on color fundus photography is essential for large-scale glaucoma screening. However, existing deep learning models are typically data-driven and lack explicit integration of retinal anatomical knowledge, which limits their robustness across heterogeneous clinical datasets. Moreover, pathological cues in fundus images may appear beyond predefined anatomical regions, making fixed-region feature extraction insufficient for reliable diagnosis. To address these challenges, we propose a retinal knowledge-oriented glaucoma screening framework that integrates dynamic multi-scale feature learning with domain-specific retinal priors. The framework adopts a tri-branch structure to capture complementary retinal representations, including global retinal context, structural features of the optic disc/cup, and dynamically localized pathological regions. A Dynamic Window Mechanism is devised to adaptively identify diagnostically informative regions, while a Knowledge-Enhanced Convolutional Attention Module incorporates retinal priors extracted from a pre-trained foundation model to guide attention learning. Extensive experiments on the large-scale AIROGS dataset demonstrate that the proposed method outperforms diverse baselines, achieving an AUC of 98.5% and an accuracy of 94.6%. Additional evaluations on multiple datasets from the SMDG-19 benchmark further confirm its strong cross-domain generalization capability, indicating that knowledge-guided attention combined with adaptive lesion localization can significantly improve the robustness of automated glaucoma screening systems.

Problem

• glaucoma screening
• fundus image
• retinal anatomical knowledge
• pathological cues
• cross-domain generalization

Innovation

• Dynamic Window Mechanism
• Knowledge-Enhanced Attention
• Retinal Anatomical Priors
• Multi-Level Feature Integration
• Adaptive Lesion Localization