Retinal Structure Detection in OCTA Image via Voting-Based Multitask Learning.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Automated detection of retinal structures like retinal vessels (RV), foveal avascular zone (FAZ), and retinal vascular junctions (RVJ) is crucial for diagnosing eye diseases.
• Current methods often struggle with the complexity of microvasculature in OCTA images, especially for precise localization and classification of RVJs.

Purpose of the Study:

• To propose a novel Voting-based Adaptive Feature Fusion multi-task network (VAFF-Net) for joint segmentation, detection, and classification of RV, FAZ, and RVJ in OCTA.
• To develop an adaptive feature fusion mechanism for enhanced multi-task learning in retinal image analysis.

Main Methods:

• Developed VAFF-Net incorporating a task-specific voting gate module for adaptive feature extraction and fusion at multiple levels.
• Designed a specialized task head combining heatmap regression and grid classification for RVJ localization and classification.
• Utilized three en face angiograms from different retinal layers, unlike single-layer approaches.

Main Results:

• Extensive experiments on three OCTA datasets demonstrated VAFF-Net's superior performance compared to state-of-the-art single-task and multi-task methods.
• The method showed generalization capabilities across different imaging modalities, including color fundus photography.
• Three new datasets for multiple structure detection were created and partially released with code.

Conclusions:

• VAFF-Net offers a robust and effective solution for multi-task learning in retinal image analysis.
• The proposed approach advances automated detection of critical retinal structures, supporting clinical decision-making.
• The generalizability and released resources contribute to further research in ophthalmic imaging analysis.