Robust multi-view approaches for retinal layer segmentation in glaucoma patients via transfer learning

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible blindness, characterized by progressive retinal nerve tissue deterioration and peripheral vision loss.
• Early diagnosis of glaucoma is crucial for preventing blindness.
• Ophthalmologists assess glaucoma progression by measuring retinal layer thickness using optical coherence tomography (OCT) scans.

Purpose of the Study:

• To develop and evaluate two novel approaches for multi-region segmentation of retinal layers in OCT images from glaucoma patients.
• To enable automated extraction of anatomical structures essential for glaucoma assessment across diverse OCT scan patterns.

Main Methods:

• Two deep learning-based segmentation approaches were developed, utilizing transfer learning for robust retinal layer segmentation.
• The first approach employs a single segmentation module for all OCT scan patterns (circumpapillary circle, macular cube, optic disc radial).
• The second approach utilizes view-specific modules, automatically selecting the appropriate one for each scan pattern.

Main Results:

• Both approaches demonstrated satisfactory performance in segmenting retinal layers across different OCT scan types.
• The first approach achieved a Dice coefficient of 0.85±0.06, excelling in radial scans.
• The second, view-specific approach achieved a Dice coefficient of 0.87±0.08, performing best on circle and cube scans.

Conclusions:

• This work presents the first multi-view segmentation of retinal layers for glaucoma patients using machine learning.
• The findings highlight the potential of AI-driven systems in assisting the early diagnosis and management of glaucoma.