OphFusionNet: Uncertainty-Driven Multi-Scale Multimodal Feature Fusion Network for Ophthalmic Diseases Classification

Area of Science:

• Ophthalmology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computer Vision

Background:

• Multimodal imaging is crucial in ophthalmology, but current automated diagnostics underutilize complementary data.
• Effective integration of diverse imaging modalities remains a challenge for accurate disease diagnosis.

Purpose of the Study:

• To develop a novel multimodal learning framework, OphFusionNet, for enhanced automated ophthalmic disease diagnosis.
• To improve multimodal data integration by addressing feature redundancy and modality dominance.

Main Methods:

• Proposed OphFusionNet, a framework featuring uncertainty-driven multi-scale multimodal feature fusion.
• Introduced a multi-scale feature fusion module with sparse self-attention (MSFF-SSA) for hierarchical representation.
• Implemented an uncertainty-aware multimodal fusion module with game-theoretic selection (UMF-GTSS) and a modality distillation strategy (MDS).

Main Results:

• OphFusionNet achieved superior multimodal integration and state-of-the-art performance on four ophthalmic datasets.
• The MSFF-SSA module enhanced feature expressiveness and efficiency.
• The UMF-GTSS and MDS components improved robustness, trustworthiness, and individual modality discriminative capacity.

Conclusions:

• OphFusionNet effectively integrates multimodal ophthalmic data for advanced automated diagnosis.
• The proposed fusion strategies significantly improve diagnostic accuracy and reliability in ophthalmology.
• This framework offers a promising direction for AI-driven ophthalmic disease detection.