A comparison of the accuracy of the original-mirror alignment algorithm and a landmark-independent method in constructing craniofacial midsagittal plane in patients with facial deformities

Objective To compare the accuracy of the original-mirror alignment algorithm and a landmark-independent method in constructing the midsagittal plane (MSP) of the cone beam computed tomography in patients with facial deformities, so as to provide a theoretical basis for symmetric analysis. Methods The study was approved by the hospital ethics committee. Cone beam computed tomography data of 30 patients with facial deformities were obtained, and the output was saved in DICOM format. The scan data were imported into Mimics 21.0; after segmentation, three-dimensional (3D) skull models were reconstructed. Furthermore, the 3D scan data of skulls were transformed into mirror skull models using Geomagic Studio 2014 reverse engineering software. The MSP of each skull was generated using both the original-mirror alignment algorithm and the landmark-independent method. Original-mirror alignment algorithm: the original skull model and its mirror model were combined, and the new data to calculate the MSP (S1) of the original data in Geomagic Studio 2014 were obtained. Landmark-independent method: the following anatomical landmarks were determined using Mimics 21.0: nasion (N), crista galli (CG), sella (S), basion (Ba), vomer (V), posterior nasal spine (PNS), incisive foramen (IF), and anterior nasal spine (ANS). The MSP (S2) of best fit was then found by minimizing the mean square distance of these eight anatomical landmarks to a plane in Geomagic Studio 2014. The results of the S1 and S2 models constructed using the original-mirror alignment algorithm and the landmark-independent method, respectively, were scored subjectively by five senior maxillofacial surgeons, and a paired t-test was performed for the two groups. The internal consistency analysis was performed based on secondary experiments to verify the repeatability of the expert evaluation method. Results The average scores of the S1 and S2 models were 65.73 and 75.90, respectively. The average score of the model constructed using the landmark-independent method was significantly higher than that of the model constructed using the original-mirror alignment algorithm (P