Improved accuracy of mandible geometry reconstruction at the stage of data processing and modeling

The article presents a comparative study of influence of the Lanczos resampling filter on improving the accuracy of reconstruction of mandible geometries. The research was performed on eight different patients. Digital Imaging and Communications in Medicine data were obtained on the Siemens Somatom Sensation Open 40 scanner. At the stage of reconstruction, the same parameters were utilized, while only slice thickness was changed. Modeling with voxel dimensions of 0.4 mm × 0.4 mm × 1.5 mm was chosen as the gold standard over the modeling approach comprising voxel dimensions of 0.4 mm × 0.4 mm × 3.0 mm and improved using the Lanczos resampling filter. The influence of the Lanczos resampling filter on the accuracy of reconstruction of mandible geometry is very similar for the eight presented patients. The average results show a distribution with a positive skew and kurtosis. The value of skewness is 0.713 and kurtosis is 4.221 for the model without Lanczos filtering applied. When the Lanczos filtering is applied the value of skewness is 0.542 and kurtosis is 4.313. Based on 95% confidence, changes in layer thickness from 1.5 mm to 3 mm generated errors reconstructing the geometry of the mandible at the value of 0.153 mm ± 1.209 mm. In models improved using the Lanczos resampling filter, the errors generated in reconstructing the geometry of the mandible were minimized at the value of 0.160 mm ± 1.007 mm. The presented research highlights new opportunities to improve the accuracy of reconstruction geometry of the mandible at the stage of data processing.