This is a retrospective cross-sectional study of patients with TED who were evaluated at the Sheba Medical Center Thyroid clinic between 2017 and 2025. Inclusion criteria were age ≥18 years, a confirmed diagnosis of TED, documented follow-up in the TED clinic after 2017, and availability of high-resolution orbital CT imaging suitable for quantitative analysis obtained prior to orbital decompression surgery, if decompression was subsequently performed. The scan used for quantitative analysis was the earliest high-resolution orbital CT available after the initial diagnosis of TED. Patients with prior orbital decompression or inadequate imaging quality were excluded.

Demographic variables (age, sex, systemic diseases and smoking status), thyroid disease–related parameters (disease duration and treatment history), and ophthalmic evaluation including the presence of strabismus, deviation angle and eye movement deficit were retrieved from the electronic medical records. Eye involvement was determined in the TED multidisciplinary Clinic by factors including exophthalmos, limitation of gaze, imaging and clinical assessment. Restrictive eye movements were defined as limitation of motility ( $\ge 2$ ) documented at orthoptics examination. The CAS score was divided into low CAS (0-2) and high (CAS (3-7), reflecting disease activity.

All CT scans were acquired with a slice thickness of 0.6 mm or less. Image processing and segmentation were performed at the PlanNet—3D Medical Solutions Center, Sheba Medical Center, using Mimics software (Materialize, Leuven, Belgium). Orbital boundaries were manually delineated between orbital bones ( Figure 1 ). Eyelids were excluded by submitting 4 pixels which are equal to 6 mm in average. After measuring Orbital volume, the orbital fat was measured by using Threshold-based segmentation.

Flowchart of study TED patients included in the FMR analysis.

3D Model: For each orbit, fat volume, globe volume (calculated from its diameter) and a constant optic nerve/sheath volume of 500 mm³ (calculated by orbital optic nerve length- 30 mm, mean diameter 4.8) were subtracted from the total orbital volume (serves as the numerator) to isolate extraocular muscle volume (serving as the denominator).

The 3D fat-to-muscle ratio (FMR₃D) was calculated as:

2D Model: On a standardized first coronal slice posterior to the globe, fat and muscle areas were segmented. The optic nerve was approximated as a constant 30 mm², yielding the 2D fat-to-muscle ratio (FMR₂D):

Both right and left orbits were analyzed separately. For primary analyses, the “more affected eye” (lower FMR) was designated as the study eye, while the contralateral orbit was analyzed separately as the “less affected eye.”

Continuous variables were summarized as mean ± SD (SD). Categorical variables are presented as counts and percentages. Between-group comparisons used parametric or nonparametric tests as appropriate, while χ² or Fisher’s exact tests were applied for categorical outcomes. Pearson and Spearman correlation was used to examine the relationship between continuous variables. Logistic regression was applied to model binary outcomes as a function of continuous or categorical predictors. Agreement between 2D and 3D FMR measurements was assessed using the intraclass correlation coefficient (ICC, two-way mixed, absolute agreement) to evaluate whether a single slice could substitute for full volumetric segmentation. We used Youden’s J statistics, to select FMR cutoffs for presence of strabismus. Model discrimination was assessed by receiver operating characteristic (ROC) analysis, reporting the area under the curve (AUC). All analyses were performed using SPSS version 28 (IBM Corp., Armonk, NY), with statistical significance defined as P <.05.