Quasi-multimodal-based pathophysiological feature learning for retinal disease diagnosis.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Multimodal data aids retinal disease diagnosis but faces challenges like data heterogeneity and complexity.
• Current methods struggle with integrating diverse ophthalmic imaging data effectively.

Purpose of the Study:

• To propose a unified framework for synthesizing and fusing multimodal retinal data for disease classification and grading.
• To enhance the accuracy and efficiency of retinal disease screening using integrated imaging techniques.

Main Methods:

• Developed a framework integrating fundus fluorescein angiography (FFA), multispectral imaging (MSI), and saliency maps.
• Employed parallel models for modality-specific representation learning and adaptive feature calibration.
• Utilized visualizations for interpreting the learning system in image and feature spaces.

Main Results:

• Achieved superior performance in multi-label classification (F1-score: 0.683, AUC: 0.953) and diabetic retinopathy grading (Accuracy: 0.842, Kappa: 0.861).
• Demonstrated the effectiveness of the proposed approach over state-of-the-art methods on public datasets.

Conclusions:

• The proposed framework offers a scalable solution for multimodal data augmentation in medical imaging.
• This work significantly improves retinal disease screening accuracy and efficiency.