Subjective Assessment of Eight CT Features in the Prediction of Dysthyroid Optic Neuropathy.

Purpose: To evaluate the subjective assessment of 8 orbital CT features for predicting dysthyroid optic neuropathy (DON).

Methods: Scan montages from 137 orbits without DON and 121 with DON were each graded independently by 3 observers for 8 imaging features: namely, degree of apical crowding, extraocular muscle enlargement, expansion of orbital fat, clarity of the superior orbital fissure, fat prolapse through the superior orbital fissure, medial wall bowing, general orbital vascular congestion, and dilation of the superior ophthalmic vein. Gradings were analyzed individually and also averaged across observers. Associations with DON were assessed using univariate and multivariate logistic regression, and performance of the latter was assessed using area under the receiver-operating-characteristic curve, sensitivity, specificity, and calibration plots.

Results: Unadjusted models showed apical crowding (p < 0.001), extraocular muscle enlargement (p < 0.001), vascular congestion (p = 0.001), and medial wall bowing (p < 0.001) were each associated with DON, across all observers. Models for observers 1 and 3 included only apical crowding, with a 1-unit grading-increase giving, respectively, a 2.88-fold and 3.04-fold increase in DON odds (p < 0.001). The observer 2 model included extraocular muscle enlargement (odds ratio: 2.95; p < 0.001) and vascular congestion (odds ratio: 2.57; p = 0.023). The "averaged-score" model for all observers included only apical crowding, with a 3.76-fold increased odds per unit increase in grading (p < 0.001). The models showed borderline-acceptable discrimination (area under the receiver-operating-characteristic curve: 0.68-0.75) and good calibration, with the averaged-score model performing best.

Conclusions: Apical crowding, extraocular muscle enlargement, vascular congestion, and medial wall bowing are key predictors of DON, with apical crowding being the most influential.