Automated Orbital MRI Yields Population Reference Data
Key Takeaways
- In 28,779 adults in the German National Cohort, automated orbital magnetic resonance imaging (MRI) morphometry produced age- and sex-stratified reference values for orbital tissues and geometry.
- Validation against expert manual segmentations suggested strong automated performance for larger orbital structures, with lower agreement mainly in smaller or thinner tissues.
- The reference dataset included benchmark measurements for total non-bony orbital volume, globe dimensions, proptosis, orbital rim angle, and optic nerve and extraocular muscle compartments.
- Males had larger measurements than females across all reported orbital parameters, and those differences were consistently statistically significant.
- Lens volume increased with age in both sexes, while vitreous volume declined slightly and orbital, optic nerve, and extraocular muscle volumes changed more modestly.
Investigators performed a cross-sectional baseline MRI analysis within the population-based German National Cohort (NAKO), as reported in Scientific Reports. From an initial MRI cohort, multilevel quality control yielded a final analytic sample of 28,779 participants. Imaging used 3-dimensional T1-weighted magnetization-prepared rapid gradient-echo (MPRAGE) brain MRI on five identical 3-T Siemens scanners with 1.0 mm isotropic resolution. A fully automated pipeline segmented 15 orbital structures and generated 34 volumetric and geometric parameters, with expert-in-the-loop ground truth expanded to 322 training orbits from 161 participants and a held-out validation set of 20 orbits from 10 participants. Deep Neural Patchworks, a U-Net-like 3D patch-based framework, was used with high-level boundary refinement and morphometric post-processing.
Validation was strongest for larger structures and anchored use of the automated approach in the reference dataset. Mean Dice scores were 0.97 for vitreous, 0.89 for lens, and 0.85 for the intraorbital optic nerve. Measurement agreement was also high for selected volumetric outputs, with intraclass correlation coefficients (ICC) of 0.92 for the total non-bony orbit and 0.96 for the vitreous. Agreement was lower for smaller or thinner structures, and Bland-Altman analysis showed a small systematic boundary bias.
In the reference dataset, mean total non-bony orbital volume was 34.4 cm³ and mean axial length was 23.5 mm. Reference values were also reported for globe dimensions, proptosis, orbital rim angle, interzygomatic distance, lens, vitreous, anterior chamber, optic nerve compartments, and combined extraocular muscles. Males had larger measurements across all parameters; total orbital volume averaged 36.6 cm³ in males and 31.7 cm³ in females, with all 34 sex comparisons reported as p < 0.001. Across age strata, lens enlargement was the clearest pattern, including an approximately 40% increase in male lens volume from ages 20 to 25 years to older than 65 years, while vitreous volume declined slightly and orbital, optic nerve, and extraocular muscle volumes changed more modestly.
The morphometry came from a standardized brain MRI protocol rather than dedicated high-resolution orbital imaging, which the authors said may reduce precision for smaller structures such as the oblique muscles. Gold-standard ophthalmic measurements and physician-confirmed diagnoses were not available, and the cohort covered adults aged 20 to 74 years. Because the analysis was cross-sectional, it did not establish causality or within-person change, and the reference ranges do not extend to pediatric or highly elderly populations. The model was also trained on T1-weighted MPRAGE scans from identical 3-T Siemens systems, so performance on other vendors, field strengths, contrast types, or severe orbit-distorting pathology remains unproven.
The investigators concluded that this analysis established a validated automated approach for orbital MRI morphometry and produced age- and sex-stratified reference measurements for orbital tissues and geometry in a large German cohort. They framed the dataset as a foundation for future high-throughput epidemiologic and morphometric research.
Clinician Questions
Which MRI protocol and scanner setup were used to generate these orbital reference data?
The reference data came from 3-dimensional T1-weighted brain MPRAGE MRI acquired at 1.0 mm isotropic resolution on five identical 3-T Siemens MAGNETOM Skyra scanners in the German National Cohort. Because this was a standardized brain MRI protocol rather than dedicated orbital imaging, the dataset is most directly applicable to similarly acquired examinations.
Which orbital structures were included in the automated segmentation pipeline?
The pipeline segmented the anterior chamber, lens, vitreous, intraorbital and intracranial optic nerve segments, optic chiasm, the six extraocular muscles, and the complete non-bony orbit including the medial and lateral orbital rims. Those segmentations supported both tissue-volume measurements and geometric reference values.
How were age-related orbital changes characterized in the German National Cohort MRI analysis?
Age-stratified percentile curves showed continuous lens enlargement in both sexes, slight vitreous decline, stable anterior chamber volume, and modest increases in orbital, optic nerve, and combined extraocular muscle volumes. As one numeric example, male lens volume increased by approximately 40% from ages 20 to 25 years to older than 65 years.
How far do these orbital MRI reference values extend beyond the studied cohort?
The reference values apply to adults aged 20 to 74 years in a German population cohort and were derived from T1-weighted MPRAGE on identical 3-T Siemens systems. The cross-sectional design does not show longitudinal change, and performance on other scanner vendors, field strengths, contrast types, pediatric populations, highly elderly populations, or severe orbit-distorting pathology was not established.