Comparative Assessment of CT and MRI in Orbital Measurement: Accuracy, Agreement, and Reliability

Background: This study aimed to compare orbital linear dimensions and volumes obtained from computed tomography (CT) and magnetic resonance imaging (MRI) in healthy adults using harmonized anatomic landmarks, and to assess inter-modality agreement, predictive accuracy, and inter-observer reliability. Methods: This retrospective study included 40 adults (80 orbits) who underwent both cranial CT and MRI (2017–2025) without orbital trauma. Orbital boundaries were defined by standardized bony landmarks. Anteroposterior (AP), transverse (TR), and craniocaudal (CC) diameters were measured on CT and T2-weighted MRI, and orbital volume was calculated using the ellipsoid formula (0.52 × AP × TR × CC). Paired statistical tests were used to compare modalities; agreement was assessed with correlations (Pearson’s r, Spearman’s ρ), predictive accuracy with regression (R²), and inter-observer reliability with intraclass correlation coefficients (ICC). Results: AP diameters showed no significant difference between modalities (R1 p = 0.595; R2 p = 0.348). TR, CC, and volumes were larger on MRI (all p ≤ 0.003). Correlations were excellent for R1 (all r ≥ 0.967; p < 0.001) but weaker for R2, especially TR (ρ = 0.219; p = 0.051). Bland–Altman plots showed minimal bias for R1 but wider limits for R2. Regression confirmed excellent MRI-to-CT prediction for R1 (R² up to 0.994) but poor for R2 (TR R² = 0.008). Inter-observer reliability was parameter-dependent: CC showed the best agreement (CT ICC = 0.651; MRI ICC = 0.724), TR the weakest, and volumes moderate-to-good. Conclusions: MRI closely approximates CT for AP diameters and provides acceptable estimates for TR, CC, and volume under standardized conditions. However, MRI yields slightly higher values and shows greater observer sensitivity, particularly for TR. CT remains the reference for orbital morphometry, whereas MRI is a promising radiation-free adjunct when combined with harmonized protocols and (semi-)automated segmentation.