Addressing Glaucoma Structure-Function Relationship: A Multi-Task Learning Framework With Multi-Modal and Unpaired...

Area of Science:

• Ophthalmology
• Neuroscience
• Medical Imaging

Background:

• Glaucoma is an irreversible neurodegenerative disease causing vision loss.
• Visual field (VF) tests are standard but time-consuming and variable.
• Optical coherence tomography (OCT) offers objective, reproducible data for glaucoma assessment.

Purpose of the Study:

• To develop a deep learning framework for predicting visual field (VF) test results using OCT data.
• To address limitations of existing methods in predicting single VF indicators and handling severe glaucoma.
• To leverage multi-modal and multi-task learning for improved VF prediction accuracy.

Main Methods:

• A multi-modal, multi-task deep learning framework utilizing OCT data.
• A dynamic weighted loss function to enhance prediction in severe glaucoma.
• A novel PairMatcher model to augment unpaired VF data.

Main Results:

• The proposed framework demonstrated superior performance compared to existing methods.
• Improved prediction accuracy for visual field indicators, including in severe glaucoma cases.
• Effective augmentation of unpaired VF data using the PairMatcher model.

Conclusions:

• The developed framework shows significant potential for accurate VF prediction in glaucoma.
• This approach can enhance clinical care by providing objective and reliable functional damage assessment.
• The multi-modal, multi-task learning strategy effectively addresses challenges in glaucoma VF prediction.