Page 41 - Advanced concepts in orbital wall fractures

was one of the first to compare a three-dimensional (3D) reconstructive
software program to a gold standard for direct measurements23. A filling
method (alleged gold standard in that study) was compared to a 3D software
program. They reported no significant volume discrepancy between the C two methods, but the accuracy was not acceptable compared to findings 2 in the recent literature. The mean volume difference was 0.93 cc (SD 1.08)
and therefore insufficient for clinical use considering how this relates to enophthalmos. Regensburg et al. compared direct measurement with a CT-based method in Mimics version 9.11 to measure bony orbital volume and orbital fat/muscle volume21. This was performed on a single phantom and showed a difference of −0.7 and +0.7 % in fat and −1.5 and −2.2 % in muscle volume compared to the known volume. No statements were made on the total bony orbital volume of the phantom. Intraobserver variability was <5 % for the calculations of fat volume, muscle volume, and bony orbital volume. This represents approximately 1.5 cc of total orbital volume, which can be considered a substantial measuring error. Strong et al. published very small intraoperator and interoperator errors when using Maxillo software5. However, comparison with a gold standard is lacking, so it is impossible to know if the real volume was measured.
In Method A, the built-in automatic segmentation was not accurate enough, probably due to the many morphological challenges hindering accurate segmentation. Method A was easy, fast, and reproducible. However, it often overestimated the volume as it frequently included parts of the surrounding bone, air (frontal/ethmoidal sinus), and inferior orbital fissure in the segmentation. This resulted in a mean difference of 0.49 cc (SD 0.74) compared to the gold standard. Therefore, this auto- segmentation is not advisable in a clinical setting.
In the newly developed method SA, bone and air masks were created. The method was designed to solve the problem of overestimation due to inclusion of bone and air in the segmentation. This resulted in higher accuracy, while the time needed to perform the segmentation increased only slightly. Mean difference compared to the gold standard was 0.24 cc (SD 0.27) and average time 146 s (SD 16.0). The SA method was still perfectly reproducible, because the creation of the mask is not observer dependent.
Volume segmentation method
39