Volumetric analysis of orbital fat and extraocular muscles in TED patients vs. normal controls using a novel fully automated atlas-based segmentation method.

Abstract

Purpose : We aim to demonstrate the use of a novel atlas-based segmentation method for orbital soft tissues on normal and thyroid eye disease (TED) patients, and to characterize differences between normal and TED orbits using this method.

Methods : A novel consensus orbital soft tissue atlas was created by iteratively aligning 32 normal individual orbit CT scans with a variety of nonlinear registration protocols and then manually refined. The atlas was then used to identify soft tissue components of normal and TED orbits using a nonlinear registration process. Manual segmentation was used to assess accuracy. T-test was used to compare the two cohorts for each extraocular muscle and orbital fat compartment volumes, as well as to compare the ratio of fat to EOM volume for TED and normal control patients.

Results : Accuracy as determined by the Jaccard index (Intersection over Union, IoU) varied between 0.705 and 0.918 while the mean Dice coefficient ranged from 0.801 to 0.952. Orbit volumes based on 32 normal orbit CT scans and 37 TED orbit scans are summarized in Table 1.
TED orbits demonstrated significantly higher fat (p = 0.00013) and total EOM volume (p = 0.00021) compared to normal orbits. The ratio of fat to EOM volume was calculated for TED and normal control patients, and the distribution plotted on Figure 1. T-test was used to compare the two cohorts. Of 32 normal orbits compared to 63 TED orbits, the fat to EOM ratio measured 3.37 and 3.36, respectively (p = 0.82).

Conclusions : This automated atlas-based process produces accurate segmentation of orbital soft tissue compartments in normal and TED patients. This is the first fully automated segmentation of orbital soft tissue components, allowing for rapid determination of specific soft tissue compartment volume and density on CT, which has many potential clinical and research applications.
There does not seem to be a bimodal distribution of fat to EOM ratio, in contrast to previous publications which have suggested that there are subtypes of TED with primarily fat hypertrophy or primarily muscle hypertrophy. While there may be outliers on either end of the spectrum, the majority of TED patients in our cohort have a similar degree of fat and muscle volume expansion, and a similar ratio compared to normal controls.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

Table 1. Normal and TED Orbit Values

 

Figure 1. Ratio of Fat to EOM Volume