The reliability of different methods of manual volumetric segmentation of pharyngeal and sinonasal subregions
The purpose of the study was to test the intra and interobserver reliability of manual volumetric segmentation of pharyngeal and sinonasal airway subregions.Study Design.
Cone beam computed tomography data of 15 patients were collected from an orthodontic clinical database. Two experienced orthodontists independently performed manual segmentation of the airway subregions. Four performance measures were considered to test intra and interobserver reliability of manual segmentation: (1) volume correlation, (2) mean slice correlation, (3) percentage of volume difference, and (4) percentage of nonoverlapping voxels.Results.
Intra and interobserver reliability was observed to be greater than 0.96 for the entire pharyngeal and sinonasal airway sinus subregions by both observers using the volume correlation method. Mean slice correlation was found to be greater than 0.84, showing the existence of nonoverlapping voxels. Therefore, the percentage of nonoverlapping voxels was used as a reliability measure and was found to be less than 20% for both intra and interobserver markings.Conclusions.
The mean slice correlation and percentage of nonoverlapping voxels were the most reliable performance measures of segmentation correctness. Volume correlation and the percentage of volume difference were observed to be the most reliable performance measures for volume correctness.