Disc-Aware Ensemble Network for Glaucoma Screening From Fundus Image.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible vision loss.
• Current automatic screening methods rely on segmentation accuracy, potentially missing visual features.
• There is a need for more robust and accurate glaucoma screening techniques.

Purpose of the Study:

• To develop a deep learning technique for direct glaucoma screening from fundus images.
• To propose a novel disc-aware ensemble network integrating global and local image features.
• To improve the accuracy and reliability of automatic glaucoma screening.

Main Methods:

• A novel disc-aware ensemble network was developed for automatic glaucoma screening.
• The network integrates four deep streams: global image, segmentation-guided, local disc region, and disc polar transformation.
• Outputs from different streams were fused for the final screening decision.

Main Results:

• The proposed method demonstrated superior performance compared to existing state-of-the-art algorithms.
• Experiments were conducted on two glaucoma datasets (SCES and SINDI).
• The disc-aware ensemble network effectively utilizes hierarchical context for improved screening.

Conclusions:

• The novel deep learning approach offers a promising alternative for glaucoma screening.
• Integrating global and local features enhances the accuracy of glaucoma detection.
• This method addresses limitations of traditional segmentation-based approaches.