Online Only
S50.e1
Measuring the airway in 3 dimensions: A reliability and accuracy study
Hakan El a  and Juan Martin Palomo b  Ankara, Turkey, and Cleveland, Ohio
Introduction: The aim of the study was to compare the reliability and accuracy of 3 commercially avail-able digital imaging and communications in medicine (DICOM) viewers for measuring upper airway volumes. Methods: Thirty cone-beam computed tomography scans were randomly selected, and the upper airway volumes were calculated for both oropharynx and nasal passage. Dolphin3D (version 11, Dolphin Imaging & Management Solutions, Chatsworth, Calif), InVivoDental (version 4.0.70, Anatomage, San Jose, Calif), and OnDemand3D (version 1.0.1.8407, CyberMed, Seoul, Korea) were compared with a previously tested manual segmentation program called OrthoSegment (OS) (developed at the Department of Orthodontics at Case Western Reserve University, Cleveland, Ohio). The measurements were repeated after 2 weeks, and the intraclass correlation coefficient was used for the reliability tests. All commercially available programs were compared with the OS program by using regression analysis. The Pearson correlation was used to evaluate the correlation between the OS and
the automatic segmentation programs. Results: The reliability was high for all programs. The highest correlation found was between the OS and Dolphin3D for the oropharynx, and between the OS and InVivoDental for nasal passage volume. A high correlation was found for all programs, but the results also showed statistically significant differences compared with the OS program. The programs also had inconsistencies among themselves. Conclusions: The 3 commercially available DICOM viewers are highly reliable in their airway volume calculations and showed high correlation of results but poor accuracy, suggesting systematic errors. (Am J Orthod Dentofacial Orthop 2010;137:S50.e1-S50.e9)
T
he upper airway has long been an area of in-terest in orthodontics, with topics such as the relationships between facial type and airway, airway shape and volume with growth and develop-ment, and the clinician’s potential to modify the air-way.1-9 However, most studies evaluating the airway have been conducted with 2-dimensional (2D) ceph-alograms, providing limited data such as linear and angular measurements, for a complex 3-dimensional (3D) structure.
With the introduction of cone-beam computed tomography (CBCT), the 3D diagnosis of the pa-tient became more accessible in dentistry. CBCT has
become a well-accepted oral and maxillofacial diag-nostic imaging technique in a short time, and this was mainly due to lower radiation exposure and shorter scan acquisition times necessary to obtain an accept-able image compared with conventional computed tomography scans.10-12 CBCT technology allows the segmentation and visualization of hollow structures such as the airway in 3 dimensions. Thus, with 3D im-aging, we are moving from lengths and angles toward volumes and surface areas.13-17 To visualize a CBCT scan, digital imaging and communications in medi-cine (DiCOM) viewer software is necessary. DiCOM is the accepted file format for a medical image, and a DiCOM viewer allows viewing, measuring, segment-ing, and complete analysis of a CBCT scan.
To segment and structure the airway means to de-lineate and remove all other surrounding structures for a clearer analysis and visualization. The segmen-tation of the airway can be done either manually or automatically. Manual segmentation requires the op-erator to delineate the airway slice by slice and then render all data into a 3D volume for analysis.13,14 Au-tomatic segmentation can be done by differentiating structures with different density values. This means that, because the airway is radiolucent, the density values for the airway are lower than the values for the surrounding soft tissues, allowing easy and automatic
0首付买车a
Clinical instructor, Department of Orthodontics, School of Dental Medicine, Hacettepe University, Ankara, Turkey.b
Associate professor and program director, Department of Orthodontics, and Director of the Craniofacial imaging Center, School of Dental Medicine, Case Western Reserve University, Cleveland, Ohio.
Partially funded by The Scientific and Technological Research Council of Turkey (TUBiTAK).
The authors report no commercial, proprietary, or financial interest in the prod-ucts or companies described in this article.
Reprint requests to: Juan Martin Palomo, Case Western Reserve University, School of Dental Medicine, 10900 euclid Ave, Cleveland, OH 44106; e-mail, palomo@case.edu.
Submitted, August 2009; revised and accepted, november 2009.0889-5406/$36.00
Copyright © 2010 by the American Association of Orthodontists.doi:10.1016/j.ajodo.2009.11.010
S50.e2 and Palomo
好冷的朋友圈说说心情
American Journal of Orthodontics and Dentofacial Orthopedics往事只能回味歌词
April 2010
was not clear or fully contained in the image and im-ages containing artifacts were excluded. each patient signed a consent form allowing the use of orthodontic records for research purposes.
The study sample consisted of 30 randomly se-lected patients. All DiCOM files were loaded to a computer with a Xeon (intel, Santa Clara, Calif) pro-cessor, running Windows XP Professional (Microsoft, Redmond, Wash) operating system. For the commer-cially available DiCOM viewers, the programs tested were Dolphin3D (version 11, Dolphin imaging & Management Solutions, Chatsworth, Calif), inVivo-Dental (version 4.0.70, Anatomage, San Jose, Calif), and OnDemand3D (version 1.0.1.8407, CyberMed, Seoul, Korea). Dolphin3D (D3D), inVivoDental (iVD), OnDemand3D (OD3D), and a custom-written program with Visual C++ (Micro s oft) for the Depart-ment of Orthodontics, School of Dental Medicine, at Case Western Reserve University called OrthoSeg-ment (OS) were used to render the nP and OP airway volumes separately.
Amirlak et al 18 used the OS program to test the reliability and accuracy of CBCT images. They used a water displacement technique for comparing the CBCT volumes with actual volumes and found that the manually segmented volumes of the OS program were highly accurate. Based on their results, we used the values from the OS program as the gold standard to which the other results were compared.
For the OP volume, the superior and inferior lim-its were slightly modified from the limits used by Ogawa et al.14 OP volume was defined as the volume of the pharynx between the palatal plane (AnS-PnS) extending to the posterior wall of the pharynx and the plane parallel to the palatal plane that passes from the most anteroinferior point of the second cervical verte-brae (Fig 1). The inferior limit of the nP airway was defined as the superior limit of the OP airway, and the superior limit was defined as the last slice before the nasal septum fused with the posterior wall of the phar-ynx. So, the superior border of the nP was defined on the axial slice first and then it was reflected to the sag-ittal plane (Fig 2). The described airway volumes were rendered with the D3D, iVD, and OD3D programs, according to their manufacturers’ recommendations (Figs 3-5).19-21 in the OS program, first, the limits of the airway were defined on the midsagittal slice on the sagittal view, and then the airway was painted slice by slice on the axial images.14 After we painted all slices between the defined limits, we  rechecked the image for any inconsistencies from the coronal and sagittal aspects. Finally, the painted images were rendered, and a 3D volume was obtained (Fig 6).
differentiation. The density values are called Houn-sfield units (HU). Automatic segmentation of the airway is significantly faster and more practical than manual segmentation, but the reliability and the ac-curacy of the method with commercially available programs have never been tested.
The aim of this study was to compare the reliabil-ity and accuracy of 3 commercially available DiCOM viewers for measuring upper airway volume. Upper airway volume was divided into oropharynx (OP) and nasal passage (nP) for this study.
MatERIal and MEtHodS
Our experimental protocol was approved by the Case Western Reserve University institutional review board, and all records used in this study were ob-tained from the patient data base of the Department of Orthodontics. All CBCT images were acquired with a Hitachi CB Mercuray Scanner (Hitachi Medi-cal Systems America, Twinsburg, Ohio) as a routine part of the initial diagnostic records for orthodontic patients. All images were pre e xisting and taken by using 2mA, 120kVp, and 12-in field of view (F mode) setting. each patient’s image data consisted of 512 slices, with a slice thickness of 0.377 mm, a reso-lution of 1024 x 1024 pixels, and 12 bits per pixel (4096 gray scale). Scans in which the defined airway
Fig 1. The superior (ANS-PNS line) and inferior (the most anteroinferior point of the second cervical verte-brae) limits of the OP volume.
American Journal of Orthodontics and Dentofacial Orthopedics El and Palomo S50.e3
Volume 137, Number 4, Supplement 1
measurements for OP airway volume was found with the iVD program (8.77 mm 3), and the highest was for the OD3D program (97.2 mm 3). The lowest mean dif-ference between the first and second measurements for the nP airway volume was found with the OD3D (70.26 mm 3), and the highest was found with the D3D program (515.78 mm 3). nP airway volume differences were generally larger compared with the OP airway volume differences.
The results of the paired samples t  test and the Pearson correlation coefficients are given on Table ii. The highest correlations were found between the OS and D3D pro-grams for OP airway, and between the OS and iVD for nP airway volume. Although there were high correlations between all programs, the results for each program showed statistically significant differences (Table ii).
dISCuSSIon
There are currently more than 15 third-party DiCOM viewers mainly for orthodontics, implantol-ogy, and oral and maxillofacial surgery.22 Although the reliability and accuracy of CBCT machines have been evaluated, testing the reliability of CBCT-relat-ed software has not gone further than error studies with a few samples.23-30
yamashina et al 31 used a soft-tissue equivalent phan-tom to evaluate the reliability and accuracy of CBCT in measuring the density values of air, water, and soft tissues. They concluded that the measurement of the air space surrounded by soft tissues was quite accurate, concluding that the airway volume acquired from CBCT is nearly a 1-to-1 representation of the real volume.in this study, 3 commercially available software programs that use automatic segmentation to calculate
All data were collected and volume render-ings performed by 1 operator (H.e.). The render-ings wer
e made separately for the nP and OP volumes. For reliability purposes, the images were remeasured 2 weeks after the first measurements. A total of 480 volume renderings were made, and all data were im-ported and organized by using excel (Microsoft). Separate spreadsheets were used for the first and sec-ond measurements in order to be blinded from the previous results. SPSS software (version 17.0, SPSS, Chicago, ill) was used for statistical analysis.
in order to check the reliability of the first and sec-ond measurements, intraclass correlation coefficient (iCC) values were used. Differences between the pro-grams were tested by using linear regression with the OS value as the constant, considering a significance of P  <0.05. To evaluate the correlation between the manual and automatic segmentation programs, the Pearson correlation coefficient was used.
RESultS
The descriptive characteristics and intraclass cor-relation coefficients of the study population are given in Table i. The reliability was high for all programs (Table i). The mean OP airway volumes were consis-tently larger than the nP airway volumes. The means of the first and second measurements were highest for the D3D program for both OP and nP airways (7444.37 ± 3250.08 a
nd 6617.50 ± 2696.69 mm 3, re-spectively). The means of the first and second mea-surements for the OP and nP airway volumes were lowest for the OD3D program (4603.06 ± 1741.03 and 3959.49 ± 1878.57 mm 3, respectively). The low-est mean difference between the first and second
Fig 2. a , Determination of the last axial slice before the nasal septum fuses with the posterior wall of the pharynx; B , its reflection on the sagittal slice for defining the superior limit of the NP airway. The inferior limit of the NP airway is defined as the palatal plane (ANS-PNS line).
什么是执行力S50.e4 El and Palomo
American Journal of Orthodontics and Dentofacial Orthopedics
April 2010
Fig 3. Airway volume rendering with the D3D program: a , OP; B , NP .
(range of density values displayed) to obtain a solid airway volume (Fig 4).
The automatic segmentation of the OD3D program requires the operator to first identify landmarks directly on the image, representing the density range of the area to be explored. The program does not allow control of the threshold for the area identified. Figure 5 shows a render-ing made with the OD3D program. even though a solid volume is visible on the 3D view, it can also be observed on the axial, sagittal, and coronal slices that the program
airway volumes were tested. even though all 3 pro-grams use automatic segmentation, this does not mean that they all use the same methods. Some differences between these programs are described below.
The iVD program allows more control where the user can “sculpt out” the desired airway volume from the rest of the 3D structures and, by adjusting the brightness and opacity values, clean out the unwanted voxels before calculating the final airway volume. The program also lets the user change the threshold values
American Journal of Orthodontics and Dentofacial Orthopedics El and Palomo S50.e5
Volume 137, Number 4, Supplement 1
示范性高中
Fig 4. Airway volume rendering with the IVD program: a , OP; B , NP .
All programs showed lower reliability when measur-ing nP volumes than for OP volumes. The nP airway is more challenging. With a manual segmentation program, it is harder to define nP areas, where the nasal turbinate and the concha region create an intricate anatomy. This area was also a challenge for the automatic segmentation programs, with most programs missing the nasal turbinates region where the nP airway has narrow spaces.
All images used in this study were preexisting, taken at 2 mA and 120 kVp during a 9.6-second rota-tion. These settings were chosen to follow the “as low
sometimes fails to render some parts of the airway, leaving empty spaces that it does not allow to be filled after display. As a result, the volume calculated is probably lower than it should be.
The D3D and iVD programs give more control by allowing the user to increase or decrease the thresh-old values. This sometimes backfires because filling an empty space in the airway by increasing the den-sity range displayed can result in an overflow of the volume into another region (Fig 3). This happened more often when calculating nP volumes.
S50.e6 El and Palomo
American Journal of Orthodontics and Dentofacial Orthopedics
April 2010
Fig 5. Airway volume rendering with the OD3D program: a , OP; B , NP .车辆保养周期
would be more accurate, maintaining the already reliable  values shown.
Manual segmentation seems to be the method with the greatest accuracy and allows the most operator control. But manual segmentation is also significant-ly more time-consuming and impractical for clinical use; it takes approximately 1 hour for both OP and nP volume calculations, whereas the same procedure with automatic segmentation can often be done in less than 5 minutes. All automatic segmentation programs compared with the manual segmentation program
as reasonably achievable” principle. Although it was shown that an image with diagnostic quality can be ob-tained with those settings, the images can sometimes be “grainy” compared with those taken at higher settings.10 These “grains” appear radiopaque and influence volume capture by not being captured in the threshold defined for the airway. A lower-quality image can also have blurry edges and not allow a sharp or clear separation between structures, somewhat interfering with the accuracy of the value collected. it is possible that, if the images used in this study were taken at higher settings, the programs