TECHNO BYTES
Effect of JPEG2000 compression on landmark identification of lateral cephalometric digital radiographs Ahmad Abdelkarim,a Pirkka Nummikoski,b Peter Gakunga,c John P. Hatch,d and S. Brent Doveb Jackson, Mississippi, and San Antonio, Tex Introduction: As digital imaging improves and digital cephalometric radiography becomes more prevalent, the need for digital storage space and transmission speed will increase. Compression of the image files is 1 method to overcome transmission overload. However, compression could compromise image quality. The purpose of this study was to determine the range of compression ratios, by using the JPEG2000 standard, within which the identification of landmarks on cephalometric radiographs is not compromised. Methods: Ten lateral cephalometric digital images were used. Six raters identified 19 landmarks under controlled viewing conditions. The images included the original uncompressed TIFF image and the JPEG2000 format at 3:1, 12:1, 50:1, and 110:1 compression ratios. The images were randomized and displayed with image processing software. The x and y coordinates of each landmark were recorded. Results: All compression ratios performed equally well compared with the original images with the exception of A-point and nasion at 110:1 and gonion at 3:1 compression ratios. All landmark identifications were precise with the exception of the maxillary incisal apex and edge at the 12:1 and 50:1 compression ratios, respectively. Conclusions: JPEG2000 is a reliable file format that can be implemented in orthodontic practice. (Am J Orthod Dentofacial Orthop 2010;138:518-24)
D
igital radiography has been used for a long time in medicine. It was not until the 1980s that the first intraoral sensors were applied in dentistry. The development of cost-effective extraoral digital technology, coupled with increased use of computers in orthodontic practice, has made direct digital cephalometric imaging a valid opportunity.1 As digital imaging increases, the need for storage space and transmission speed also increases. Digital imaging systems will continue to improve, and, as a result, the storage and transmission requirements for the images they produce will probably increase as well. One method to overcome this transmission overload is to compress the image files. Image compression requires less storage space
From the Schools of Dentistry, University of Mississippi Medical Center and University of Texas Health Science Center at San Antonio. a Assistant Professor, Department of Care Planning and Restorative Sciences, University of Mississippi College of Dentistry. b Professor, Oral and Maxillofacial Radiology Division, Department of Dental Diagnostic Science, UTHSCSA. c Assistant professor, Department of Orthodontics, UTHSCSA. d Professor, Departments of Psychiatry and Orthodontics, UTHSCSA. The authors report no commercial, proprietary, or financial interest in the products or companies described in this article. Reprint requests to: Ahmad Abdelkarim, Department of Care Planning and Restorative Sciences, 2500 North State Street, Jackson, MS 39216-4505; e-mail,
[email protected]. Submitted, July 2008; revised and accepted, February 2009. 0889-5406/$36.00 Copyright Ó 2010 by the American Association of Orthodontists. doi:10.1016/j.ajodo.2009.02.029
518
and decreases the transmission time, since the compressed file is smaller because of the reduced amount of binary data used to represent the image.2 There are 2 methods of image compression: lossless and lossy. Lossless compression eliminates nonessential information in the image while conserving essential data so that the digital image can be reconstructed exactly.3 Lossy compression, on the other hand, although offering considerably higher compression ratios and smaller file sizes, involves irreversible loss of data that could be essential.4 The most common file format that offers lossy compression is the joint photographic experts group (JPEG) format. The JPEG2000 file format was developed to address some deficiencies in the original JPEG standard. This file format has been adopted by the digital imaging and communication in medicine (DICOM) standard.5 Compared with JPEG, JPEG2000 allows higher compression without compromising quality. It also offers progressive image reconstruction, provides more options and greater flexibility than the standard JPEG format, and offers an optional lossless compression mode.6 Typically, the greater the compression applied to the image to reduce its file size, the lower the image quality. With JPEG2000, there is no artifact with high compression. Several studies in dentistry have addressed the effects of compression on the diagnosis of caries,7,8
American Journal of Orthodontics and Dentofacial Orthopedics Volume 138, Number 4
periapical lesions,9 subtraction radiography,10 and endodontic file length assessment.11 However, only 2 of these were done with the JPEG2000 standard.8,10 The only study that evaluated the effect of compression on lateral cephalometric images used an aluminum test block object to assess the effect of the JPEG compression algorithm on direct digital cephalometric image quality.12 It was concluded that JPEG compression had no effect on the perceptibility of landmarks in the aluminum test object used in that study. The purpose of this study was to determine the range of compression ratios, using the newly established JPEG2000 standard, within which landmark identification on cephalometric radiographs is not compromised. MATERIAL AND METHODS
Thirty lateral cephalometric digital images were obtained from the MiPACS Dental Enterprise Solution archiving system (version 3.1, Medicor Imaging, Charlotte, NC). These images had been acquired for routine orthodontic treatment. All radiographic images were acquired with a ProMax (Planmeca Oy, Helsinki, Finland) radiographic machine and exported in uncompressed tagged image file format (TIFF). Those radiographs were selected based on the following inclusion criteria: sharpness of the image, ultimate brightness and contrast, minimal noise, and full visualization of all relevant anatomic structures. From this data set, 10 images were randomly selected. The inherent resolution of these images was 254 pixels per inch, which is equivalent to 100 pixels per centimeter (1 pixel 5 0.1 mm). Image matrix dimension was 2272 3 1818 pixels at 8-bit depth. This resulted in a file size of 4.15 MB. The images were not enhanced in any way. Photoshop CS2 software (Adobe Systems, San Jose, Calif) was used for compression. Photoshop does not include the JPEG2000 plug-in by default, so it was installed manually. Photoshop uses a quality-factor scale that ranges from 1 (lowest quality) to 100 (highest quality) for JPEG2000 compression. Four JPEG2000 compressed image groups were created by using this software at quality factors of 100, 25, 6, and 1. This produced image file sizes of 1.38 MB, 0.35 MB, 83 KB, and 37.7 KB, respectively. This resulted in compression ratios of 3:1, 12:1, 50:1, and 110:1. The compression ratio expresses the difference between the file size of the original image and the file size of the same image when compressed. The quality factor of 100 was included in the study because of the reduction in size of the file (one third) and because it was the highest-quality factor that can be
Abdelkarim et al
519
achieved by the software. The quality factor of 1 was chosen because it was the lowest quality factor. An example of compressed radiographs is shown in Figure 1 at the mandibular symphysis area. Note in these images how compression at higher levels tends to lose the trabecular bony structure. Another example is shown in Figure 2 at the frontonasal suture and ridge of the nose. Note how soft tissues tend to be affected more than hard tissues by higher compression ratios. To evaluate the effect of compression at different levels, a cephalometric image at 3:1 compression ratio was qualitatively subtracted from its corresponding uncompressed TIFF image. University of Texas Health Science Center at San Antonio ImageTool software version 3.0 (available on the Internet at http://ddsdx. uthscsa.edu/dig/itdesc.html) was used for this purpose. The resultant subtracted image indicated that no pixel data had been altered from the original TIFF image. Therefore, the JPEG2000 3:1 compression ratio is lossless. On the other hand, subtraction performed for JPEG2000 images at 12:1, 50:1, and 110:1 compression ratios from the original TIFF showed pixel changes of 59.9%, 69.3%, and 70.2%, respectively. This confirmed that compression at these levels is lossy. According to Figure 3, most of the pixels that correspond to the patient’s head were changed. It seems that the majority of the 29.8% of unchanged pixels were those that corresponded to air around the patient’s head; these do not change by compression because they are uniformly black and can be compressed efficiently. Because the ImageTool software used for viewing images does not support the JPEG2000 format, all JPEG2000 images were converted to TIFF. This should not affect the quality of the images; it neither improves nor degrades them. Another subtraction experiment was performed and confirmed this. The result was 50 images consisting of the original uncompressed images and 4 compression ratios. Randomization of the images was achieved by using a randomizer to blind the viewers. The images were viewed on an OptiPlex 745 computer (Dell, Round Rock, Tex) with a Core 2 Duo Processor (Intel, Santa Clara, Calif) and the Windows XP professional operating system (Microsoft, Redmond, Wash). The video card was Graphics Media Accelerator 3000 (Intel) using up to 256 MB of system memory. The images were displayed on an UltraSharp 24-in 2007FP TFT flat panel (Dell). Screen resolution was 1920 3 1200 at 60 Hz and 16.7 million colors. The images were presented at 50% zoom because the screen resolution was smaller than the images themselves. Each image was displayed by using the ImageTool software. For each image, the viewers were asked to
520
Abdelkarim et al
American Journal of Orthodontics and Dentofacial Orthopedics October 2010
Fig 1. Compressed radiographs at 3:1, 12:1, 50:1, and 110:1 compression ratios zoomed at the mandibular symphysis and incisor. The uncompressed radiograph is identical to the compressed radiograph at 3:1 compression ratio. Note that at the higher compression ratio (right), the trabecular bony structure tends to disappear.
Fig 2. Compressed radiographs at 3:1, 12:1, 50:1, and 110:1 compression ratios zoomed at the frontonasal suture and ridge of nose. Note that at the higher compression ratios (right), the frontonasal suture tends to disappear. It seems that soft tissues such as the ridge of the nose suffer more from higher compression. The effect of compression on soft-tissue landmarks was not investigated in this study.
identify 19 landmarks commonly identified by orthodontists and previously used in research: N, Or, ANS, PNS, A, B, Pog, Gn, Me, Go, Ar, Po, S, UIA, UIE, LIE, LIA, UM, LM.13-18 Definitions of these landmarks are listed in Table I.19 Six viewers consisting of postdoctoral orthodontic and oral and maxillofacial radiology residents were carefully instructed to identify the landmarks on each image. They were guided to use the computer cursor to position
the arrow over the landmark and left click with the mouse to indicate the selection. The cursor in ImageTool software is an accurate pointed cursor. The viewers were reminded that it was important that they choose the landmarks in the same order from 1 to 19. They were allowed to correct the points after placement if needed. No time limits were set for the viewing sessions. After each landmark was chosen, the ImageTool software automatically registered the x and y coordinates of
Abdelkarim et al
American Journal of Orthodontics and Dentofacial Orthopedics Volume 138, Number 4
521
Table I. Cephalometric landmarks used in the study and their definitions
Fig 3. Subtracted compressed radiograph at 110:1 from the original (70.2% of pixels turned white, corresponding to pixels that changed in value). Almost all pixels that correspond to the patient’s head have been changed. The 29.8% of unchanged pixels are those that mostly correspond to air around the patient’s head, and these do not typically change by the compression algorithm because they are uniformly black and can be compressed efficiently.
the pixel that was selected, and a green point was positioned over the selected landmark. Transmission time over a network is directly proportional to the file size. An experiment to test transmission time over the Internet of the uncompressed image file and the 4 compressed image files was performed. The same internet speed was used for the experiment at 54 Mb per second. For the compressed files at 3:1, 12:1, 50:1, and 110:1 compression ratios, the time required for uploading or downloading was about 35%, 12%, 6%, and 3% of the time required for uploading or downloading the uncompressed files, respectively.
1
N
2
Or
3
ANS
4
PNS
5
A
6
B
7
Pog
8
Gn
9
Me
10
Go
11
Ar
12
Po
13 14 15 16 17 18 19
S UIA UIE LIE LIA UM LM
Statistical analysis
The null hypotheses were that there is no difference among the tested compression ratios in identification of landmarks with respect to landmark location, measurement precision, or rater agreement on landmark identification. Analyses were carried out by using SPSS software (version 16.0.1, SPSS, Chicago, Ill), and statistical significance was defined at the 2-sided P \0.05 level. 1.
The hypothesis that JPEG2000 compression introduces bias in the measurement of landmark location was tested by comparing the mean pixel locations in x and y spaces as a function of compression level
2.
Nasion: the point in the skull where the nasal and frontal bones unite Orbitale: the lowest point on the inferior margin of the orbit Anterior nasal spine: the tip of the anterior nasal spine Posterior nasal spine: the tip of the posterior spine of the palatine bone, at the junction of the hard and soft palates A-point: the innermost point on the contour of the premaxilla between ANS and the incisor B-point: the innermost point on the contour of the mandible between the incisor and the bony chin Pogonion: the most anterior point on the contour of the chin Gnathion: the center of the inferior point on the mandibular symphysis Menton: the most inferior point on the mandibular symphysis Gonion: the midpoint of the contour connecting the ramus and body of the mandible Articulare: the point of intersection between the shadow of the zygomatic arch and the posterior border of the mandibular ramus Porion: the midpoint of the upper contour of the metal ear rod of the cephalometer (machine porion) Sella: the midpoint of the cavity of sella turcica Upper incisor apex Upper incisor edge Lower incisor edge Lower incisor apex Mesiobuccal cusp of the upper first molar Mesiobuccal cusp of the lower first molar
by using multivariate analysis of variance (MANOVA). The analysis model contained 1 fixed between-subjects effect with 6 levels corresponding to the 6 raters, 1 repeated-measures effect with 5 levels corresponding to the original TIFF, and 4 compressed radiograph images and the interaction between raters and compression. If the omnibus main effect test for compression level was significant at the P\0.05 level, then Bonferroni protected pairwise comparisons were used to compare each level of JPEG2000 compression vs the TIFF image (k 5 4 comparisons). The hypothesis that the precision (ie, the spread of values around the mean value) of the measurements varies as a function of the level of JPEG2000 compression was tested by using a t test for the difference between variances from 2 correlated populations. Each level of JPEG2000 compression was compared with the TIFF image, and the resulting
522
Abdelkarim et al
American Journal of Orthodontics and Dentofacial Orthopedics October 2010
Mean difference (in millimeters) between the original TIFF and each compression ratio (CR) for the x coordinate
Table III. Mean difference (in millimeters) between the original TIFF and each compression ratio (CR) for the y coordinate
Landmark
Landmark
Table II.
3:1 CR
12:1 CR
50:1 CR
110:1 CR
Significance
0.10 0.06 0.12 0.04 0.11 0.01 0.02 0.14 0.06 0.43* 0.04 0.04 0.04 0.17 0.05 0.01 0.14 0.03 0.05
0.65 0.00 0.23 0.25 0.18 0.06 0.12 0.08 0.29 0.31 0.00 0.06 0.00 0.02 0.00 0.01 0.05 0.08 0.12
0.01 0.30 0.39 0.35 0.18 0.03 0.01 0.04 0.04 0.22 0.02 0.08 0.05 0.03 0.04 0.05 0.08 0.14 0.09
0.19 0.09 0.20 0.53 0.46* 0.07 0.05 0.04 0.16 0.14 0.11 0.02 0.07 0.22 0.11 0.06 0.11 0.07 0.09
0.049* 0.034* 0.400 0.000* 0.004* 0.280 0.340 0.200 0.230 0.008* 0.520 0.500 0.510 0.200 0.500 0.590 0.630 0.790 0.047*
N Or ANS PNS A B Pog Gn Me Go Ar Po S UIA UIE LIE LIA UM LM *P \0.05.
3.
P values were Bonferroni adjusted within families of comparisons (k 5 4). The hypothesis that interrater agreement varies as a function of JPEG2000 compression was evaluated by using the intraclass correlation coefficient.
RESULTS
The mean differences in millimeters between the location of the original TIFF image and that of each compressed image are presented for each landmark in Tables II and III. These mean differences were taken over the 6 viewers and the 10 radiographs. The mean differences in millimeters between the variance of the original TIFF image and that of each compressed image are presented in Tables IV and V. Pairwise comparisons demonstrated that only the following 3 landmarks had statistically significant differences from the original TIFF, which was considered the gold standard: A-point at the 110:1 compression ratio was significantly anterior by 0.46 mm with respect to the uncompressed image at the x coordinate, Go at the 3:1 compression ratio was significantly anterior by 0.43 mm with respect to the uncompressed image at the x coordinate, and N at the 110:1 compression ratio was significantly inferior by 0.39 mm with respect to the uncompressed image at the y coordinate. Testing the precision of landmark identification, we found that none of these values was statistically significant with the exception of the upper incisal apex (UIA)
N Or ANS PNS A B Pog Gn Me Go Ar Po S UIA UIE LIE LIA UM LM
3:1 CR
12:1 CR
50:1 CR
110:1 CR
Significance
0.19 0.03 0.15 0.03 0.16 0.22 0.08 0.01 0.02 0.35 0.28 0.03 0.04 0.25 0.01 0.04 0.17 0.10 0.01
0.32 0.17 0.16 0.17 0.17 0.23 0.04 0.26 0.11 0.24 0.12 0.02 0.10 0.30 0.02 0.04 0.02 0.05 0.07
0.17 0.15 0.28 0.12 0.07 0.09 0.10 0.14 0.02 0.20 0.07 0.06 0.02 0.28 0.10 0.02 0.05 0.02 0.08
0.39* 0.06 0.28 0.18 0.01 0.27 0.14 0.30 0.13 0.06 0.07 0.01 0.05 0.12 0.02 0.00 0.11 0.05 0.09
0.034* 0.810 0.098 0.350 0.490 0.076 0.392 0.076 0.110 0.066 0.120 0.280 0.710 0.050 0.140 0.560 0.860 0.410 0.052
*P \0.05.
that was statistically different (t value was 3.2) at the 12:1 compression ratio vs the original TIFF, and the upper incisal edge (UIE) that was statistically different (t value was 3.5) at the 50:1 compression ratio vs the original TIFF. Testing rater agreement, we found that all values were high (P .0.05). All intraclass correlations were uniformly high, ranging from 0.77 to 0.99 and indicating excellent agreement among the 6 raters. DISCUSSION
Most of the pixels that corresponded to the patient’s head had changed with compression. The numbers 59.9%, 69.3%, and 70.2% that correspond to pixel change in the studied compression ratios are arguable because they correspond to the whole image. The appropriate percentage would be much higher and perhaps closer to 100%, showing that most pixels corresponding to biologic information have been altered. This qualitative subtraction corresponds to the number of pixels affected by compression. The number of pixels affected is not the sole factor, though. It does not represent how much these pixels were affected. Our hypothesis was that lossy compression with JPEG2000 does not cause visually detectable changes in the radiographs. Only 2 landmarks demonstrated statistical significance at the x coordinate from the original TIFF image. The mean location of Go was statistically significant by 0.46 mm anteriorly at the 3:1 compression ratio. By
Abdelkarim et al
American Journal of Orthodontics and Dentofacial Orthopedics Volume 138, Number 4
Variance difference (in millimeters) between the original TIFF and each compression ratio (CR) for the x coordinate
Table V.
Landmark
Landmark
Table IV.
N Or ANS PNS A B Pog Gn Me Go Ar Po S UIA UIE LIE LIA UM LM
3:1 CR
12:1 CR
50:1 CR
110:1 CR
0.26 0.78 3.04 2.49 0.15 0.08 1.55 1.07 1.74 0.08 0.16 0.13 0.21 0.61 0.07 0.04 2.70 2.76 0.67
0.14 0.87 3.51 0.41 0.46 0.02 0.81 1.02 0.13 1.29 0.60 0.35 0.09 2.70* 0.33 0.05 2.55 1.55 0.03
0.06 1.61 1.94 1.18 0.55 0.13 1.52 0.11 0.44 0.15 1.14 0.01 0.18 0.26 14.35* 0.57 0.45 1.37 2.51
0.31 0.42 2.21 0.02 0.51 0.19 0.57 0.24 2.44 2.00 0.57 0.36 0.36 1.04 0.35 0.21 1.94 0.21 1.10
523
Variance difference (in millimeters) between the original TIFF and each compression ratio (CR) for the y coordinate
N Or ANS PNS A B Pog Gn Me Go Ar Po S UIA UIE LIE LIA UM LM
3:1 CR
12:1 CR
50:1 CR
110:1 CR
0.50 1.33 0.50 0.24 0.49 2.47 0.70 1.09 0.20 1.10 0.37 0.24 0.25 1.57 1.29 1.13 1.68 1.60 0.45
0.45 1.17 1.85 0.82 1.41 2.02 1.88 1.95 0.48 0.13 0.53 0.39 0.30 1.81 0.24 0.62 1.65 1.18 1.28
0.44 5.62 3.26 1.19 0.44 3.18 1.34 1.36 0.29 0.26 0.69 0.19 0.28 3.69 0.77 0.41 1.00 1.07 0.45
0.14 0.09 3.71 0.93 0.98 0.27 1.46 4.13 0.20 0.54 0.34 0.16 0.46 3.85 0.50 0.19 1.90 0.54 0.18
*Statistically significant.
using the subtraction test, it was previously demonstrated that a compression ratio of 3:1 is lossless, implying that images at this compression level are identical to the original uncompressed images. Therefore, this statistical finding is consistent with type I error. For A-point, the mean of all viewers was statistically significant at the 110:1 compression ratio for the horizontal axis. The distance between the means was 0.46 mm in the anterior direction. A discrepancy of this magnitude is considered clinically minimal. Furthermore, regions with a gradual curve, such as A-point and Go, render a more difficult task in identification, and errors tend to be proportionately greater.13 Vertically, only N demonstrated a statistically significant mean at the 110:1 compression ratio from the original TIFF image in our study. It is arguable that a higher compression ratio causes the frontonasal suture to disappear, making identification difficult. Interestingly, Stabrun and Danielsen14 found that N had high measurement error in the x coordinate. For this landmark, the distance between the means was 0.39 mm in the inferior direction. Testing the reliability of landmark identification, Baumrind and Frantz13 found a disquieting number of gross errors and outliers associated with N identifications. Thus, it is most likely that these 2 findings represent random error as well. Therefore, in general, landmark identification was accurate at all compression ratios. Testing the precision of landmark identification at different compression ratios demonstrated statistical
significance for 2 landmarks only in the horizontal axis. These landmarks were related to 1 structure (the maxillary incisor) at the incisal apex and edge at the 12:1 and 50:1 compression ratios, respectively. Testing the validity of landmark identification, Tng et al20 found that, for the maxillary incisor apex, the mean differences along both axes were statistically significant and the standard deviations were large. On the other hand, it was found that the mean differences were smaller for the maxillary incisor edge and so were the standard deviations. Also, precision is considered good in identification of the maxillary incisal edge, where the edge folds sharply.13 Furthermore, it was expected that any significance would be noted at a higher compression, but these 2 findings did not confirm that this most likely represents type I error. Thus, overall, landmark identification was precise at all compression ratios. The fact that there was high accuracy, precision, and raters’ agreement in landmark identification at all compression ratios might be explained by the following 3 factors: (1) high accuracy: the minimal deterioration of the images resulting from compression at any level is unlikely to change viewers’ identification of any landmark; (2) high precision: if poor judgment is inherent in the rater, then changing the image (regardless of quality) will not affect the location of the landmark being identified; and (3) raters’ agreement: viewers’ expertise combined with the choice of standard cephalometric landmarks routinely used in orthodontics explains the strong agreement between raters.
524
Abdelkarim et al
Cephalometric analysis requires identifying specific landmarks and calculating various angular and linear dimensions. Our study was limited to landmark identification with no investigation of actual angular and linear calculations. Performance of actual measurements that orthodontists make on a compressed image might or might not increase the discrepancy in the final outcome. After all, cephalometric radiographs are not used solely for landmark identification. We speculated that the outcome of any angular and linear measurements on a compressed cephalometric image would not be clinically significant. Cephalometric images include hard-tissue and softtissue landmarks. Based on previous research of the accuracy and precision of landmark identification, we attempted to choose the same hard-tissue landmarks. Orthodontists make angular and linear measurements that usually do not include soft-tissue landmarks; we speculated that soft tissues suffer more with compression, but the effect of compression on these landmarks is still unknown. Moreover, compression of noisy images with poor brightness or contrast would perhaps further deteriorate soft-tissue landmarks. Finally, the highest compression ratio that can be reached with the software used (110:1) was not enough to significantly deteriorate the accuracy and precision of landmark identification. The original radiographs had high resolution, another factor explaining minimal visual deterioration of the radiographs even at high compression ratios. The compression ratios in our study were limited by the software we used. Therefore, if the technology evolves, higher compression ratios should be studied. More compression would not reduce the file size significantly; therefore, it would be unnecessary. CONCLUSIONS
In light of ever-increasing network speeds and falling prices of storage media, one would aim not to reduce the file size as much as possible and to retain diagnostic information. Another valid perspective is using a file format that requires less transmission time and less storage space. In orthodontic practice, some clinicians might choose the second philosophy. Based on these statistically validated findings, JPEG2000 compression is a reliable file format that can be implemented in orthodontic practice. Changing the file format from the lossless TIFF to JPEG2000 by using the highestquality factor (least compression ratio) rendered an identical image with significant reduction in file size. High-compression ratios are negotiable, especially for preserving diagnostic value and risk management.
American Journal of Orthodontics and Dentofacial Orthopedics October 2010
REFERENCES 1. Brennan J. An introduction to digital radiography in dentistry. J Orthod 2002;29:66-9. 2. Baxes GA. Digital image processing. New York: John Wiley & Sons; 1994. 3. Lodwick GS, Taaffe JL. Radiology systems of the nineties: meeting the challenge of change. J Digit Imaging 1988;1:4-12. 4. Visser H, Rodig T, Hermann KP. Dose reduction by directdigital cephalometric radiography. Angle Orthod 2001;71: 159-63. 5. National Electrical Manufacturers Association. Digital imaging and communication in medicine (DICOM). Part 1: introduction and overview. Rosslyn, Va: National Electrical Manufacturers Association; 2006. 6. Foos D, Muka E, Slone R, Erickson B, Flynn M, Clunie D, et al. JPEG2000 compression of medical imagery. Proceedings of SPIE Digital Library 2000;3980:85-96. 7. Janhom A, van der Stelt PF, Sanderink GCH. A comparison of two compression algorithms and the detection of caries. Dentomaxillofac Radiol 2002;31:257-63. 8. Ghayyath A, Dove SB, Prihoda T, McDavid D. The effect of JPEG2000 compression on the diagnosis of interproximal caries [thesis]. San Antonio: University of Texas Health Science Center at San Antonio; 2007. 9. Koenig L, Parks E, Analoui M, Eckert G. The impact of image compression on diagnostic quality of digital images for detection of chemically-induced periapical lesions. Dentomaxillofac Radiol 2004;33:37-43. 10. Fidler A, Likar B, Pernus F, Skaleric U. Comparative evaluation of JPEG and JPEG2000 compression in quantitative digital subtraction radiography. Dentomaxillofac Radiol 2002;31: 379-84. 11. Siragusa M, McDonnell DJ. Indirect digital images: limit of image compression for diagnosis in endodontics. Int Endod J 2002;35:991-5. 12. Wenger NA, Tewson DH, McDonald F. Direct digital lateral cephalometry: the effects of JPEG compression on image quality. Med Eng Phys 2006;28:560-7. 13. Baumrind S, Frantz RC. The reliability of head film measurements. 1. Landmark identification. Am J Orthod 1971;60: 111-27. 14. Stabrun AE, Danielsen K. Precision in cephalometric landmark identification. Eur J Orthod 1982;4:185-96. 15. Macri V, Wenzel A. Reliability of landmark recording on film and digital lateral cephalograms. Eur J Orthod 1993;15:137-48. 16. Chen YJ, Chen SK, Chang HF, Chen KC. Comparison of landmark identification in traditional versus computer-aided digital cephalometry. Angle Orthod 2000;70:387-92. 17. Chen YJ, Chen SK, Yao JC, Chang HF. The effects of differences in landmark identification on the cephalometric measurements in traditional versus digitized cephalometry. Angle Orthod 2004;74: 155-61. 18. McClure SR, Sadowsky PL, Ferreira A, Jacobson A. Reliability of digital versus conventional cephalometric radiology: a comparative evaluation of landmark identification error. Semin Orthod 2005;11:98-110. 19. Miyashita K. Contemporary cephalometric radiography. Carol Stream: Quintessence; 1996. 20. Tng TT, Chan TC, Ha¨gg U, Cooke MS. Validity of cephalometric landmarks. An experimental study on human skulls. Eur J Orthod 1994;16:110-20.
