ORIGINAL ARTICLE
Accuracy Assessment of Image-Based Surface Meshing for Volumetric Computed Tomography Images in the Craniofacial Region Sang-Hoon Kang, DDS, PhD,*† Moon-Key Kim, DDS, PhD,*† Hak-Jin Kim, DDS, MSD,† Piao Zhengguo, DDS, PhD,‡ and Sang-Hwy Lee, DDS, PhD† Background: Three-dimensional printing and computer-assisted surgery demand a high-precision three-dimensional mesh model created from computed tomography (CT) imaging data using an image-based meshing algorithm. We aimed to evaluate the threedimensional geometric accuracy of surface meshes produced from CT images with commercially available software packages. Methods: The CT images were acquired for 3 human dry skulls and 10 manufactured plastic skulls. Four commercially available software packages were used to produce the surface meshes in stereolithography (STL) file format. These CT-based STL surface meshes were registered and compared with three-dimensional opticalscanned reference mesh surface for evaluating the accuracy of the STL mesh produced with each software package. Results: The surface geometries produced by the CT-image–based meshing process were all relatively accurate; differences from the three-dimensional optical-scanned data were in the voxel or subvoxel range. However, when comparisons with the three-dimensional optical-scanned surface data were performed in individual anatomic regions, we found significantly different accuracies of the CT-based STL surface meshes produced by the different software packages. Conclusions: We found that all 4 software packages showed reasonably good meshing accuracies for clinical use. However, the range of errors inherent in the CT-image–based meshing process demands that caution should be taken in selecting and manipulating the software to avoid potential errors in specific clinical applications. Key Words: CT, stereolithography, mesh, accuracy, craniofacial (J Craniofac Surg 2014;25: 2051–2055)
From the *Department of Oral and Maxillofacial Surgery, National Health Insurance Service, Ilsan Hospital, Goyang-si, Republic of Korea; †Department of Oral and Maxillofacial Surgery, College of Dentistry, Yonsei University, Seoul, Republic of Korea; ‡Department of Oral Maxillofacial Surgery, Stomatological Hospital of Guangzhou Medical College, Guangzhou, China. Received September 12, 2013. Accepted for publication June 13, 2014. Address correspondence and reprint requests to Sang-Hwy Lee, DDS, PhD, Department of Oral and Maxillofacial Surgery, College of Dentistry, Yonsei University, 50 Yonsei-ro, Seodaemun-gu, Seoul 120-752, Republic of Korea; E-mail: [email protected] Supported by the National Research Foundation of Korea grant funded by the Korean government (2012R1A1A2003829) for L.S.H. The authors report no conflicts of interest. Copyright © 2014 by Mutaz B. Habal, MD ISSN: 1049-2275 DOI: 10.1097/SCS.0000000000001139
C
omputed tomography (CT) is a major diagnostic tool in the medical imaging field. The CT imaging data provide abundant, in-depth information about body regions of interest. Threedimensional shaping of the region based on CT imaging data provides a three-dimensional understanding of biologic structures and allows biomechanic analyses of structures with the finite element method. The CT-based imaging data can also be very useful for designing and producing medical devices and models for dental implant surgery, orthognathic surgery, and reconstructive cancer surgery. The threedimensional printing or additive manufacturing technology can be used as part of the computer-assisted surgical approach.1 It is not possible to perform three-dimensional printing directly with CT data to produce biologic models or medical devices, such as surgical guides and templates. The file format necessary for three-dimensional printing is different from that of medical CT imaging, which is generally known as digital imaging and communication in medicine (DICOM). The DICOM data must be converted to a format suitable for three-dimensional printing, such as the stereolithography (STL) format. The STL format is one of several standard file types used for three-dimensional printing. The STL approximates the three-dimensional surface of the anatomic structure with triangular polygonal meshes of different sizes and shapes to achieve a smooth and exact structural representation.2 A meshing process is applied to the CT image data to create an STL surface. This process requires image-based mesh generation, which is an automated process that creates computerized models by extracting the surface information from CT data and creating the triangular polygonal meshes.2–6 Although various meshgenerating techniques are currently available, they are typically developed from computer-aided design (CAD) softwares, and therefore, they have difficulty in meshing complex biologic structures, such as craniofacial models, from volumetric medical CT data. The accuracy of the model or device produced by the threedimensional printing technology depends on the quality of the original imaging data in DICOM format, the quality of the reconstructed threedimensional digital model with segmentation, and the meshing accuracy of the surface. The DICOM data can be converted to different surface data, depending on the algorithms applied in the surface meshing process. The manufacturers of commercial software typically do not disclose the exact algorithms used for the conversion of DICOM to STL, but presumably, the algorithms would not be identical to each other. Therefore, they may have different degrees of meshing accuracy, which may or may not influence the final quality of the three-dimensional printed product and, by extension, the treatment outcome. To the best of our knowledge, no previous studies have directly reported on the degree of accuracy of different software programs for surface meshing from CT imaging data of the craniofacial skeleton. Therefore, in this study, we investigated the meshing accuracy of STL surfaces from CT imaging data with image-based
The Journal of Craniofacial Surgery • Volume 25, Number 6, November 2014
Copyright © 2014 Mutaz B. Habal, MD. Unauthorized reproduction of this article is prohibited.
2051
The Journal of Craniofacial Surgery • Volume 25, Number 6, November 2014
Kang et al
meshing software programs. We converted volumetric CT images of skulls in DICOM format to STL mesh surfaces with 4 commercially available software packages and verified their accuracies by analyzing and comparing the STL surfaces to the three-dimensional optical scanning data of the skulls.
MATERIALS AND METHODS Our strategy was to adopt 2 different models with different types of material characteristics with the aim of minimizing potential biases in accuracy due to the thresholding and segmentation process. Two types of skulls were used in the current study; group 1 was composed of 3 human dry skulls (from collections of the anatomy section, in the Department of Oral Biology, Yonsei University, Seoul, Republic of Korea), and group 2 was composed of 10 commercially available plastic skulls (A20 classic human skull model; 3B Scientific, Hamburg, Germany). There was no remarkable evidence of defects in the dry human skulls, but their mandibles were not evaluated in this study, because of the possibility of mobility during the experiments.
DICOM Data Archiving From CT The CT images were acquired for each skull in groups 1 and 2 to produce the DICOM image data sets with a Sensation 64 CT scanner (Siemens AG, Erlangen, Germany). The parameters were set to a pixel size of 0.4375 mm, a resolution of 512 512, a field of view of 22.40 cm, and a slice thickness of 0.6 mm. The CT data for each skull were saved in DICOM file format and exported to a personal computer.
Meshing of DICOM Data to STL Surface Four commonly used commercial software packages were arbitrarily selected and used to evaluate their accuracy in image-based meshing of CT images to produce STL mesh surfaces. The selected software packages were InVivoDental (version 5.0; Anatomage, San Jose, CA), Mimics (version 14.0; Materialise NV, Leuven, Belgium), OnDemand3D (APP version 1.0; CyberMed Inc, Seoul, Korea), and OsiriX Imaging Software (version 3.7; Pixmeo, Geneva, Switzerland). The DICOM CT data of 3 human dry skulls and 10 plastic skulls were imported into the software packages to create the three-dimensional image models. The STL meshing functions were used to convert the DICOM data to the STL format (Figs. 1A, B). The parameters could not be standardized for three-dimensional image reconstruction and STL meshing, because each software platform had a different array of control settings. Instead, for each software platform, we selected parameters aimed to optimize the process and quality of the three-dimensional image construction and STL conversion to attain the best possible meshing results. However, when the parameter settings could not be specified during the meshing operation, the default setting values were regarded as optimal. Individual parameter values for each software package were set as follows: 1. InVivoDental: The parameter isovalue was set to auto; the subsample value of the isosurface is 1. The smoothing category was set to a relax value of 0.1, power value of 10, and thickness value of 2. 2. Mimics: The threshold values for bones were between 226 and 3071. The predefined quality settings were considered optimal during three-dimensional modeling. Other parameters included triangle reduction of 3, edge angle of interaction of 10 degrees, and tolerance of 0.0547 mm. 3. OnDemand3D: The initial setting value was set to a compress category value of 70 and a smoothing value of 3. The threshold values were between 100 and 4000.
2052
FIGURE 1. Representative images illustrate the experiments performed for evaluating the surface distances between CT-based and optical-scanned surface meshes. A, Surface mesh data formatted in STL were constructed from (B) three-dimensional reconstructed CT image data. C, CT-based and optical-scanned STL mesh surfaces of a skull were registered and superimposed to measure the intersurface distances between mesh surfaces. D, Superimposed STL surfaces derived from CT and optical-scanning data were analyzed to determine intersurface distances over the entire surface; the intersurface distances are shown in colors that indicate the degree and direction of discrepancy. The color-coded discrepancy bar (right) shows positive values (warm colors) or negative values (cool colors) when the CT-based surface is larger or smaller, respectively, than the optical surface. E, The intersurface distances were also evaluated at the center points of different anatomic regions, delineated here and filled with different colors: (1) zygoma (blue), (2) nasal (yellow), (3) maxillary (brown/black), (4) frontal (green), (5) temporal-parietal (blue green), (6) alveolar (violet), and (7) occipital (silver blue); the regions of both sides were included in the analyses. The mandible (gold) and teeth (light blue) were omitted from the analyses.
4. OsiriX: The decimate resolution value was set to 0.01, smooth iteration value was 20, and the pixel value of the first surface category was 300.
Numbers (1–4, as shown in the list above) were randomly assigned to the software packages to avoid biases during the measurements and analyses of results.
Three-Dimensional Optical-Scanned Reference Model to be Compared With the CT-based STL Skull Models The skulls of groups 1 and 2 were optically scanned with the smartSCAN3Dduo (Breuckmann, Meersburg, Germany), and the data were stored in STL format. The resolution of the scanner camera was 1.3 megapixels, and the accuracy range was within ±15 μm. Rapidform XOV2 (INUS Technology, Seoul, Korea) was used for comparing the optical-scanned reference STL surface and the STL surface derived from meshing the DICOM data (Fig. 1C). The optical-scanned and meshed STL images were registered and overlapped with the minimum shortest distance, based on the iterative closest point algorithm in the XOV2 software. The distance between the optical-scanned and meshed STL images was calculated and displayed in colors by the anatomic regions, with differences color coded to the values shown in the discrepancy bar (Fig. 1D). The craniofacial anatomic points were placed at the center of each anatomic region; these included the right and left zygoma and nasal, maxillary, frontal, temporal-parietal, alveolar, and occipital points (Fig. 1E). The intersurface distance between the 2 STL surfaces, that is, the optical-scanned surface and the DICOM-based STL surface, was measured 3 times at 7 anatomic points in groups 1 (total measurements, N = 126) and 2 (total measurement, N = 420). The intersurface distances were assigned negative values when the DICOM-based © 2014 Mutaz B. Habal, MD
Copyright © 2014 Mutaz B. Habal, MD. Unauthorized reproduction of this article is prohibited.
The Journal of Craniofacial Surgery • Volume 25, Number 6, November 2014
STL was smaller than the three-dimensional optical-scanned STL, for both the surface and point measurements. The statistical significance of the intersurface distances was analyzed with the Statistical Package for the Social Sciences (version 18.0; SPSS Inc, Chicago, IL); statistical significance was calculated with analysis of variance and a post hoc test, and Ps < 0.05 were considered statistically significant. To evaluate errors related to registration and superimposition methods, 5 plastic skulls were randomly selected. Two identical STL surfaces of a skull were registered with the registration functions in the XOV2. Their intersurface distances were again measured at all anatomic points to evaluate the methods-related error.
RESULTS When we measured the intersurface distance between the three-dimensional optical-scanned and CT-based STL images, a high proportion of the color-coded distances were blue (negative values) for group 1 (images not shown) and yellow and green (positive values) for group 2 (Fig. 2). Representative intersurface distances for group 2 are shown in discrepancy map for each software package (Figs. 2A–D). The overall intersurface distances for skulls in groups 1 and 2 ranged between −0.53 and 0.74 mm (software 1), −0.72 and 0.84 mm (software 2), −0.73 and 0.86 mm (software 3), and −0.86 and 0.75 mm (software 4). The CT-based and optical-scanned STL surface data were also compared at different anatomic points to calculate the intersurface distances (Tables 1–4). For group 1 (dry skulls), the discrepancies in the image surfaces created with software packages 2 to 4 were generally greater than those created with software 1 (Table 1). For group 2 (plastic skulls), the discrepancies in the image surfaces created with software 4 were greater than those created with the other software packages (Table 3). The frontal, occipital, and temporal-parietal points showed relatively greater discrepancies than the other anatomic points for both groups 1 and 2.
CT Image-based STL Surface Meshing
TABLE 1. Intersurface Distances Between the Optical-Scanned and CT-Based STL Surfaces Rendered With Different Software Packages (1–4), Measured at the Indicated Anatomic Points for Group 1 (Human Dry Skulls) Anatomic Location Software 1 Zygoma Nasal Maxillary Frontal Occipital Temporal-parietal Alveolar Total (N = 126)
−0.17 (0.11) −0.03 (0.14) −0.05 (0.10) −0.31 (0.29) −0.14 (0.35) −0.27 (0.26) −0.12 (0.16) −0.16 (0.24)
Software 2 Software 3 Software 4 −0.37 −0.24 −0.24 −0.51 −0.19 −0.44 −0.29 −0.32
(0.13) (0.17) (0.11) (0.26) (0.21) (0.23) (0.14) (0.21)
0.30 (0.11) 0.60 (0.18) 0.45 (0.11) 0.26 (0.25) 0.56 (0.19) 0.09 (0.32) 0.32 (0.12) 0.37 (0.26)
−0.39 −0.34 −0.24 −0.56 −0.15 −0.50 −0.28 −0.35
(0.14) (0.20) (0.06) (0.33) (0.24) (0.26) (0.16) (0.25)
P*
Accuracy Assessment of Image-Based Surface Meshing for Volumetric Computed Tomography Images in the Craniofacial Region Sang-Hoon Kang, DDS, PhD,*† Moon-Key Kim, DDS, PhD,*† Hak-Jin Kim, DDS, MSD,† Piao Zhengguo, DDS, PhD,‡ and Sang-Hwy Lee, DDS, PhD† Background: Three-dimensional printing and computer-assisted surgery demand a high-precision three-dimensional mesh model created from computed tomography (CT) imaging data using an image-based meshing algorithm. We aimed to evaluate the threedimensional geometric accuracy of surface meshes produced from CT images with commercially available software packages. Methods: The CT images were acquired for 3 human dry skulls and 10 manufactured plastic skulls. Four commercially available software packages were used to produce the surface meshes in stereolithography (STL) file format. These CT-based STL surface meshes were registered and compared with three-dimensional opticalscanned reference mesh surface for evaluating the accuracy of the STL mesh produced with each software package. Results: The surface geometries produced by the CT-image–based meshing process were all relatively accurate; differences from the three-dimensional optical-scanned data were in the voxel or subvoxel range. However, when comparisons with the three-dimensional optical-scanned surface data were performed in individual anatomic regions, we found significantly different accuracies of the CT-based STL surface meshes produced by the different software packages. Conclusions: We found that all 4 software packages showed reasonably good meshing accuracies for clinical use. However, the range of errors inherent in the CT-image–based meshing process demands that caution should be taken in selecting and manipulating the software to avoid potential errors in specific clinical applications. Key Words: CT, stereolithography, mesh, accuracy, craniofacial (J Craniofac Surg 2014;25: 2051–2055)
From the *Department of Oral and Maxillofacial Surgery, National Health Insurance Service, Ilsan Hospital, Goyang-si, Republic of Korea; †Department of Oral and Maxillofacial Surgery, College of Dentistry, Yonsei University, Seoul, Republic of Korea; ‡Department of Oral Maxillofacial Surgery, Stomatological Hospital of Guangzhou Medical College, Guangzhou, China. Received September 12, 2013. Accepted for publication June 13, 2014. Address correspondence and reprint requests to Sang-Hwy Lee, DDS, PhD, Department of Oral and Maxillofacial Surgery, College of Dentistry, Yonsei University, 50 Yonsei-ro, Seodaemun-gu, Seoul 120-752, Republic of Korea; E-mail: [email protected] Supported by the National Research Foundation of Korea grant funded by the Korean government (2012R1A1A2003829) for L.S.H. The authors report no conflicts of interest. Copyright © 2014 by Mutaz B. Habal, MD ISSN: 1049-2275 DOI: 10.1097/SCS.0000000000001139
C
omputed tomography (CT) is a major diagnostic tool in the medical imaging field. The CT imaging data provide abundant, in-depth information about body regions of interest. Threedimensional shaping of the region based on CT imaging data provides a three-dimensional understanding of biologic structures and allows biomechanic analyses of structures with the finite element method. The CT-based imaging data can also be very useful for designing and producing medical devices and models for dental implant surgery, orthognathic surgery, and reconstructive cancer surgery. The threedimensional printing or additive manufacturing technology can be used as part of the computer-assisted surgical approach.1 It is not possible to perform three-dimensional printing directly with CT data to produce biologic models or medical devices, such as surgical guides and templates. The file format necessary for three-dimensional printing is different from that of medical CT imaging, which is generally known as digital imaging and communication in medicine (DICOM). The DICOM data must be converted to a format suitable for three-dimensional printing, such as the stereolithography (STL) format. The STL format is one of several standard file types used for three-dimensional printing. The STL approximates the three-dimensional surface of the anatomic structure with triangular polygonal meshes of different sizes and shapes to achieve a smooth and exact structural representation.2 A meshing process is applied to the CT image data to create an STL surface. This process requires image-based mesh generation, which is an automated process that creates computerized models by extracting the surface information from CT data and creating the triangular polygonal meshes.2–6 Although various meshgenerating techniques are currently available, they are typically developed from computer-aided design (CAD) softwares, and therefore, they have difficulty in meshing complex biologic structures, such as craniofacial models, from volumetric medical CT data. The accuracy of the model or device produced by the threedimensional printing technology depends on the quality of the original imaging data in DICOM format, the quality of the reconstructed threedimensional digital model with segmentation, and the meshing accuracy of the surface. The DICOM data can be converted to different surface data, depending on the algorithms applied in the surface meshing process. The manufacturers of commercial software typically do not disclose the exact algorithms used for the conversion of DICOM to STL, but presumably, the algorithms would not be identical to each other. Therefore, they may have different degrees of meshing accuracy, which may or may not influence the final quality of the three-dimensional printed product and, by extension, the treatment outcome. To the best of our knowledge, no previous studies have directly reported on the degree of accuracy of different software programs for surface meshing from CT imaging data of the craniofacial skeleton. Therefore, in this study, we investigated the meshing accuracy of STL surfaces from CT imaging data with image-based
The Journal of Craniofacial Surgery • Volume 25, Number 6, November 2014
Copyright © 2014 Mutaz B. Habal, MD. Unauthorized reproduction of this article is prohibited.
2051
The Journal of Craniofacial Surgery • Volume 25, Number 6, November 2014
Kang et al
meshing software programs. We converted volumetric CT images of skulls in DICOM format to STL mesh surfaces with 4 commercially available software packages and verified their accuracies by analyzing and comparing the STL surfaces to the three-dimensional optical scanning data of the skulls.
MATERIALS AND METHODS Our strategy was to adopt 2 different models with different types of material characteristics with the aim of minimizing potential biases in accuracy due to the thresholding and segmentation process. Two types of skulls were used in the current study; group 1 was composed of 3 human dry skulls (from collections of the anatomy section, in the Department of Oral Biology, Yonsei University, Seoul, Republic of Korea), and group 2 was composed of 10 commercially available plastic skulls (A20 classic human skull model; 3B Scientific, Hamburg, Germany). There was no remarkable evidence of defects in the dry human skulls, but their mandibles were not evaluated in this study, because of the possibility of mobility during the experiments.
DICOM Data Archiving From CT The CT images were acquired for each skull in groups 1 and 2 to produce the DICOM image data sets with a Sensation 64 CT scanner (Siemens AG, Erlangen, Germany). The parameters were set to a pixel size of 0.4375 mm, a resolution of 512 512, a field of view of 22.40 cm, and a slice thickness of 0.6 mm. The CT data for each skull were saved in DICOM file format and exported to a personal computer.
Meshing of DICOM Data to STL Surface Four commonly used commercial software packages were arbitrarily selected and used to evaluate their accuracy in image-based meshing of CT images to produce STL mesh surfaces. The selected software packages were InVivoDental (version 5.0; Anatomage, San Jose, CA), Mimics (version 14.0; Materialise NV, Leuven, Belgium), OnDemand3D (APP version 1.0; CyberMed Inc, Seoul, Korea), and OsiriX Imaging Software (version 3.7; Pixmeo, Geneva, Switzerland). The DICOM CT data of 3 human dry skulls and 10 plastic skulls were imported into the software packages to create the three-dimensional image models. The STL meshing functions were used to convert the DICOM data to the STL format (Figs. 1A, B). The parameters could not be standardized for three-dimensional image reconstruction and STL meshing, because each software platform had a different array of control settings. Instead, for each software platform, we selected parameters aimed to optimize the process and quality of the three-dimensional image construction and STL conversion to attain the best possible meshing results. However, when the parameter settings could not be specified during the meshing operation, the default setting values were regarded as optimal. Individual parameter values for each software package were set as follows: 1. InVivoDental: The parameter isovalue was set to auto; the subsample value of the isosurface is 1. The smoothing category was set to a relax value of 0.1, power value of 10, and thickness value of 2. 2. Mimics: The threshold values for bones were between 226 and 3071. The predefined quality settings were considered optimal during three-dimensional modeling. Other parameters included triangle reduction of 3, edge angle of interaction of 10 degrees, and tolerance of 0.0547 mm. 3. OnDemand3D: The initial setting value was set to a compress category value of 70 and a smoothing value of 3. The threshold values were between 100 and 4000.
2052
FIGURE 1. Representative images illustrate the experiments performed for evaluating the surface distances between CT-based and optical-scanned surface meshes. A, Surface mesh data formatted in STL were constructed from (B) three-dimensional reconstructed CT image data. C, CT-based and optical-scanned STL mesh surfaces of a skull were registered and superimposed to measure the intersurface distances between mesh surfaces. D, Superimposed STL surfaces derived from CT and optical-scanning data were analyzed to determine intersurface distances over the entire surface; the intersurface distances are shown in colors that indicate the degree and direction of discrepancy. The color-coded discrepancy bar (right) shows positive values (warm colors) or negative values (cool colors) when the CT-based surface is larger or smaller, respectively, than the optical surface. E, The intersurface distances were also evaluated at the center points of different anatomic regions, delineated here and filled with different colors: (1) zygoma (blue), (2) nasal (yellow), (3) maxillary (brown/black), (4) frontal (green), (5) temporal-parietal (blue green), (6) alveolar (violet), and (7) occipital (silver blue); the regions of both sides were included in the analyses. The mandible (gold) and teeth (light blue) were omitted from the analyses.
4. OsiriX: The decimate resolution value was set to 0.01, smooth iteration value was 20, and the pixel value of the first surface category was 300.
Numbers (1–4, as shown in the list above) were randomly assigned to the software packages to avoid biases during the measurements and analyses of results.
Three-Dimensional Optical-Scanned Reference Model to be Compared With the CT-based STL Skull Models The skulls of groups 1 and 2 were optically scanned with the smartSCAN3Dduo (Breuckmann, Meersburg, Germany), and the data were stored in STL format. The resolution of the scanner camera was 1.3 megapixels, and the accuracy range was within ±15 μm. Rapidform XOV2 (INUS Technology, Seoul, Korea) was used for comparing the optical-scanned reference STL surface and the STL surface derived from meshing the DICOM data (Fig. 1C). The optical-scanned and meshed STL images were registered and overlapped with the minimum shortest distance, based on the iterative closest point algorithm in the XOV2 software. The distance between the optical-scanned and meshed STL images was calculated and displayed in colors by the anatomic regions, with differences color coded to the values shown in the discrepancy bar (Fig. 1D). The craniofacial anatomic points were placed at the center of each anatomic region; these included the right and left zygoma and nasal, maxillary, frontal, temporal-parietal, alveolar, and occipital points (Fig. 1E). The intersurface distance between the 2 STL surfaces, that is, the optical-scanned surface and the DICOM-based STL surface, was measured 3 times at 7 anatomic points in groups 1 (total measurements, N = 126) and 2 (total measurement, N = 420). The intersurface distances were assigned negative values when the DICOM-based © 2014 Mutaz B. Habal, MD
Copyright © 2014 Mutaz B. Habal, MD. Unauthorized reproduction of this article is prohibited.
The Journal of Craniofacial Surgery • Volume 25, Number 6, November 2014
STL was smaller than the three-dimensional optical-scanned STL, for both the surface and point measurements. The statistical significance of the intersurface distances was analyzed with the Statistical Package for the Social Sciences (version 18.0; SPSS Inc, Chicago, IL); statistical significance was calculated with analysis of variance and a post hoc test, and Ps < 0.05 were considered statistically significant. To evaluate errors related to registration and superimposition methods, 5 plastic skulls were randomly selected. Two identical STL surfaces of a skull were registered with the registration functions in the XOV2. Their intersurface distances were again measured at all anatomic points to evaluate the methods-related error.
RESULTS When we measured the intersurface distance between the three-dimensional optical-scanned and CT-based STL images, a high proportion of the color-coded distances were blue (negative values) for group 1 (images not shown) and yellow and green (positive values) for group 2 (Fig. 2). Representative intersurface distances for group 2 are shown in discrepancy map for each software package (Figs. 2A–D). The overall intersurface distances for skulls in groups 1 and 2 ranged between −0.53 and 0.74 mm (software 1), −0.72 and 0.84 mm (software 2), −0.73 and 0.86 mm (software 3), and −0.86 and 0.75 mm (software 4). The CT-based and optical-scanned STL surface data were also compared at different anatomic points to calculate the intersurface distances (Tables 1–4). For group 1 (dry skulls), the discrepancies in the image surfaces created with software packages 2 to 4 were generally greater than those created with software 1 (Table 1). For group 2 (plastic skulls), the discrepancies in the image surfaces created with software 4 were greater than those created with the other software packages (Table 3). The frontal, occipital, and temporal-parietal points showed relatively greater discrepancies than the other anatomic points for both groups 1 and 2.
CT Image-based STL Surface Meshing
TABLE 1. Intersurface Distances Between the Optical-Scanned and CT-Based STL Surfaces Rendered With Different Software Packages (1–4), Measured at the Indicated Anatomic Points for Group 1 (Human Dry Skulls) Anatomic Location Software 1 Zygoma Nasal Maxillary Frontal Occipital Temporal-parietal Alveolar Total (N = 126)
−0.17 (0.11) −0.03 (0.14) −0.05 (0.10) −0.31 (0.29) −0.14 (0.35) −0.27 (0.26) −0.12 (0.16) −0.16 (0.24)
Software 2 Software 3 Software 4 −0.37 −0.24 −0.24 −0.51 −0.19 −0.44 −0.29 −0.32
(0.13) (0.17) (0.11) (0.26) (0.21) (0.23) (0.14) (0.21)
0.30 (0.11) 0.60 (0.18) 0.45 (0.11) 0.26 (0.25) 0.56 (0.19) 0.09 (0.32) 0.32 (0.12) 0.37 (0.26)
−0.39 −0.34 −0.24 −0.56 −0.15 −0.50 −0.28 −0.35
(0.14) (0.20) (0.06) (0.33) (0.24) (0.26) (0.16) (0.25)
P*
