Objective:

To develop and validate a publicly available computed tomography (CT) atlas of the human orbit that can serve as a spatial reference framework for future neuro-ophthalmic imaging studies.

Background:

Quantitative imaging of orbital structures remains limited by the absence of a standardized reference space. While brain imaging benefits from well-established public atlases, no analogous dataset exists for the orbit. This lack of a common template hinders reproducibility in studies involving orbital volumetrics, segmentation algorithms, and disease-related morphometrics.

Design/Methods:

16 normal orbital CT scans (9 male, 7 female; mean age 52.8 years) were retrospectively selected from clinical archives. Each image was mirrored to create 32 standardized orbits, which were iteratively aligned using rigid, affine, and nonlinear registration with Advanced Normalization Tools. The resulting averaged image was manually segmented by expert neuroradiologists and ophthalmologists to label key orbital structures. Consensus segmentations were generated using the Simultaneous Truth and Performance Level Estimation algorithm. Atlas performance was evaluated using automated atlas-based segmentation on 32 additional normal orbits and 63 orbits from patients with thyroid eye disease (TED).

Results:

Atlas-based segmentation demonstrated high agreement with manual ground truth for normal orbits (Dice coefficient 0.80–0.95) and lower, yet acceptable, agreement in TED orbits (0.61–0.82). As expected, TED orbits exhibited significantly increased extraocular muscle and fat volumes compared with controls (p < 0.001). The final dataset, including labeled atlas and description files, is publicly available for research use.

Conclusions: