STAGE challenge: Structural-Functional Transition in Glaucoma Assessment

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a primary cause of irreversible blindness, with diagnosis relying on time-consuming visual field (VF) tests.
• Current deep learning models for glaucoma often focus on structural images (OCT, fundus) but neglect functional VF outcomes.

Purpose of the Study:

• To bridge the gap between structural imaging and functional vision assessment in glaucoma.
• To establish a standardized benchmark and dataset for evaluating algorithms that predict functional VF indicators from structural OCT images.

Main Methods:

• The STAGE Challenge utilized a dataset of 401 OCT volumes with corresponding VF labels and demographic data.
• Seven finalist teams developed dual-branch deep learning architectures integrating OCT and tabular data.
• Task-specific OCT models were investigated for improved performance in predicting VF indicators.

Main Results:

• All participating teams employed dual-branch architectures combining OCT and tabular data.
• Models utilizing task-specific OCT approaches demonstrated superior performance in predicting glaucoma's functional visual field indicators.
• The challenge provided a standardized framework for comparing structure-function analysis algorithms.

Conclusions:

• Tailored deep learning strategies are crucial for accurately linking structural imaging data to functional outcomes in glaucoma assessment.
• The STAGE Challenge dataset and framework serve as a foundation for future research in glaucoma structure-function analysis.