A-eye: Automated 3D MRI segmentation and morphometric feature extraction for eye and orbit atlas construction

Area of Science:

• Medical Imaging
• Ophthalmology
• Artificial Intelligence

Background:

• Ophthalmic diagnostics and treatments require accurate anatomical data.
• Previous eye and orbit segmentation studies used small datasets and varied imaging methods.
• Standardized tools for analyzing ocular structures from Magnetic Resonance Images (MRI) are lacking.

Purpose of the Study:

• To develop and validate an automated 3D segmentation method for the healthy human adult eye and orbit using MRI.
• To create large-scale, unbiased eye atlases for standardization in ophthalmic research.
• To automate the estimation of ophthalmic morphometry biomarkers and explore correlations with body mass index.

Main Methods:

• Leveraged a large dataset of 1245 T1-weighted MRI scans.
• Employed the deep learning framework nnU-Net for MR-Eye segmentation.
• Implemented quality control protocols for reliable large-scale data segmentation.

Main Results:

• Achieved robust and accurate 3D segmentation of ocular and orbital structures (lens, globe, optic nerve, rectus muscles, orbital fat).
• Automated estimation of key ophthalmic morphometry biomarkers, including axial length and volumetry.
• Benchmarked correlations between body mass index and eye structure volumes.
• Generated the first large-scale unbiased eye atlases (female, male, combined).

Conclusions:

• The developed automated 3D segmentation pipeline provides accurate and reliable analysis of ocular and orbital structures from MRI.
• The generated large-scale eye atlases will serve as a crucial resource for standardizing spatial normalization tools in ophthalmic research.
• This work advances ophthalmic diagnostics and treatments through improved data analysis and standardization.