Objective. The objective of this study was to compare dental radiographs printed on glossy paper from calibrated low-
cost printers with monitor display.
Study design. Three typical intraoral radiographs were selected and a questionnaire was developed with questions
assessing accuracy and subjective quality. A test pattern was designed for printer calibration. After calibration,
radiographs were printed on glossy paper with 3 ink-jet and 2 thermo-sublimation printers. Sixteen raters evaluated the
printed radiographs, 9 of them also on standardized viewing monitors. Subjective ratings were compared, and an
ROC-analysis based on expert-consensus monitor readings was performed.
Results. Low inter-rater reproducibility (mean Cohen’s Kappa monitor: 0.49; printers: 0.44), but high diagnostic
accuracy was found for all printers (areas [Az] underneath the ROC curves: 0.725 to 0.884). The overlap between the
95% Az confidence intervals of the mean indicate no significant differences.
Conclusion. Our preliminary findings indicate that dental radiographs may be evaluated on glossy paper prints of
calibrated customary printers. (Oral Surg Oral Med Oral Pathol Oral Radiol Endod 2008;106:578-86)
As digital radiographic technology becomes more and dentist in an appropriate way. Since a digital file is
more commonplace, the need for simple and reliable certainly not appropriate for offices not yet equipped
transfer of digital images from one office to another digitally, alternative solutions must be found. Cost-
increases. Although today personal computers are used effective hardcopies on paper would provide a useful
throughout almost every working environment, a few alternative for this purpose.
offices still operate without any digital technology. Budget printers considered here include typical off-
These offices are not capable of managing a digital the-shelf ink-jet technology, plus thermo-sublimation
image file, even if it is encoded in typical digital image- printers belonging to the mid-range price segment. For
file formats. According to German regulations,1,2 the comparison, professional dry or wet laser radiographic
producer of a digital radiograph must provide it to a printers typically range between $25,000 and $55,000,
which is obviously out of range for a regular dental
office considering the limited number of prints. Not
a
Assistant Professor, Departments of Oral Surgery (and Oral Radi- much scientific information has been published on the
ology), Johannes Gutenberg University, Mainz, Germany. topic after introduction of a prototype ink-jet printer for
b
Head of Section for Dentomaxillofacial Radiology, Department of
printing of various radiographs by Kirkhorn et al. in
Craniomaxillofacial Surgery, Albert-Ludwigs-Universität, Freiburg,
Germany. 1994.3 The same group published encouraging results
c
Private practice, Schleswig-Holstein Dental Association, Germany. from a Receiver Operating Characteristics (ROC)
d
Private practice, Bavarian Dental Association, Germany. analysis for simulated tumors in a chest phantom.4 In
e
Head of Dental Standardization Committee, Deutsches Institut für dental radiography, we are aware of only 2 subjec-
Normung (German Institute for Standardization), Pforzheim, Ger- tive image quality studies5,6 and 2 accuracy studies
many.
f
Carestream Health, Deutschland GmbH, Stuttgart, Germany. using paper prints from customary printers.7,8 On the
g
Private practice, Germany. other hand, expensive wet or dry radiographic laser
h
Chairman of Standardization Committee, Siemens AG Medical So- imagers explicitly constructed for the production of
lutions, Erlangen, Germany. high-quality hardcopies are well established, e.g., in
Received for publication Mar 15, 2007; returned for revision Aug 7, digital mammography9 or for pulmonal nodule de-
2007; accepted for publication Sep 14, 2007.
1079-2104/$ - see front matter tection.10,11 Because of the reasons explained above,
© 2008 Mosby, Inc. All rights reserved. however, they do not fall within the scope of this
doi:10.1016/j.tripleo.2007.09.004 article.
578
OOOOE
Volume 106, Number 4 Schulze et al 579
Fig. 2. Test pattern developed for printer calibration. Test features included horizontal and vertical optical resolution (upper left
corner), color gradient (lower half and upper left corner), contrast resolution (circular frame around image plus scripture
”Bildqualität”) as well as geometric reproduction (bars in the middle).
root canal filling) as compared to the monitor mean. Table II. Kappa values for assessment of inter-rater
Here, the certainty of decision was assessed on a reproducibility
3-point confidence scale (1 ⫽ pathology definitely Monitor Print out
present, 2 ⫽ uncertain, 3 ⫽ pathology definitely not Mean 0.49 0.44
present). Another 4 major questions subdivided into 14 SD 0.14 0.14
single questions had been prepared for subjective rank- Median 0.51 0.46
ing of image quality on a 3-point scale (1 ⫽ diagnos- Minimum 0.19 0.04
Maximum 0.85 0.74
tically acceptable, 2 ⫽ uncertain, 3 ⫽ diagnostically
not acceptable). The remaining 3 questions (1.4, 3.1, Values are based on all observations per category.
and 3.3) provided a fourth answer option not following
this scale. Thus, they were not evaluated in this study.
A total of 16 observers (11 from 3 different univer- conference. This consensus reading was subsequently
sity dental schools; 5 from different private offices) taken as a ”weak gold standard” representing the
participated in the study. This sample comprised 12 ”truth” with respect to the monitor observations.
general dental practitioners, 3 oral surgeons, and 1
medical radiologist, with their working experience Statistical data analysis
ranging between 1 and 20 years. Please note that in All statistics were performed with SPSS (Version
Germany there is no specialization in dental radiology. 12.0.1) for Windows (SPSS Inc., Chicago, IL). As
Nine observers (1 radiologist, 2 oral surgeons, and 6 explained earlier, 4 questions (subdivided into 14 single
general practitioners) from 2 university dental schools questions) were used for accuracy assessment as com-
also evaluated the radiographs on 2 standardized mon- pared to the mean monitor observations. An ROC anal-
itors complying with German regulations2: 6 of them ysis was calculated, with the area (Az) beneath the
on a 17-inch cathode ray tube (CRT) monitor (Vision ROC curve indicating diagnostic accuracy.
Master Pro 413, IIyama Electric Co. Ltd., Nagano, Inter-rater agreement was assessed using Cohen’s
Japan) with an image size of 25.5 cm ⫻ 19.5 cm; the kappa. To obtain symmetrical cross-correlation tables
remaining 3 on a 19-inch liquid crystal display (LCD) required for a Kappa analysis, a positive and a negative
monitor (SCENICVIEW P19-1S, Fujitsu Siemens, Mu- rating category were computed. The undecided rating
nich, Germany; image size: 27.5 cm ⫻ 21 cm). By ”uncertain (2)” was assigned to the negative category.
German law,1 standardized viewing monitors have to The monitor evaluation is based on all observations
be available in every office using digital radiography. from the 9 observers.
These monitors have to be labeled after an initial thor-
ough evaluation performed by authorized personnel. To RESULTS
pass this evaluation, the monitors have to provide suf- The analysis of the test-pattern prints revealed that 2
ficient contrast, luminance (intensity), optical resolu- printers (CS5, THP) had difficulties with the criterion
tion, and geometric accuracy, together with additional ”color gradient,” where either stripe artifacts were pro-
requirements all detailed in the guidelines.2 A daily duced or, even worse, an inverse gradient was recog-
plus a monthly quality check by means of the well- nizable. A low-contrast detail depicted by the final
known SMPTE test pattern is mandatory. Evaluations letter of the writing “Bildqualität,” was reproduced by
were all made in a darkened, quiet room. Paper prints none of the printers. THP, however, even failed to
were viewed on a white cardboard support using indi- reproduce the last 2 letters. All other features of the
rect light. The observers were blinded with respect to pattern were printed sufficiently.
printer type and study design. Altogether, we evaluated Inter-rater reproducibility was low for this investiga-
28 ⫻ 16 ⫽ 448 observations from each of the 5 printers tion (Table II), with a mean kappa value of 0.49 (⫾
and 28 ⫻ 9 ⫽ 252 observations from the calibrated 0.14) for the monitor versus 0.44 (⫾ 0.14) for the
monitors. printout observation. A separate evaluation of the ques-
tions used for accuracy analysis revealed very similar
Establishment of a ”weak gold standard” values (printer: 0.45 ⫾ 0.16; monitor: 0.42 ⫾ 0.20). For
In addition, 2 experts in radiographic diagnostics both evaluations, the Wilcoxon test indicated no signif-
(D.S., R.S.) with a working experience of more than 10 icant difference (P ⫽ .53).
years rated the images independently on either of the 2 Interproximal caries had to be detected in 1 radio-
standardized and calibrated viewing monitors detailed graph, where 50% of the surfaces according to the
above. This was also done in a quiet dimmed-light consensus monitor evaluation either surely had a lesion
environment. In case of disagreement (50% of the or the observers were uncertain. For this question, KDP
cases), a consensus was reached during a telephone had the highest mean Az (0.893), followed by CS5
OOOOE
582 Schulze et al October 2008
priate radiograph transfer also in instances where the quality was based on the display of the periodontal liga-
recipient is not yet using digital technology. For exam- ment space, marginal bone level, and the pulp cavity.
ple in Germany, ink-jet printers typically cost between Inter-rater reproducibility was low, a drawback that is
$120 and $500, with an estimated cost per glossy-paper common knowledge in radiographic image interpreta-
print between $1 and $4. Using budget nonproprietary tion,12,13 particularly for enamel approximal lesions,14
ink can further reduce these costs. Although the overall which may have been suspected in our images. Also, the
print costs are relatively high, this has to be compared observers had not been specifically trained for the partic-
to the high expense for dry or wet laser printers com- ular radiographic tasks. To compensate for this well-
monly used in medical radiography. The quality of known handicap, we decided to use a relatively high
paper prints from digital radiographs has been assessed number of experienced observers from many centers,
by relatively few studies,3,6-8 where only7,8 are con- which also should eliminate bias due to familiarization
cerned with accuracy evaluation on dental radiographs. with center-specific image quality.10 Owing to the multi-
Two printers (CS5, EPS) used in this study were typical center design of the study, we had to use 2 viewing
off-the-shelf ink-jet printers and 1 (KDP) was a budget monitors instead of only 1. Since the initial and follow-up
semi-professional thermo-sublimation photo printer. calibration required by German regulations1,2 provides a
The remaining 2 printers (THP, CHO) were constructed high level of standardization, however, we believe that
particularly for radiograph printing. The printers were this drawback should be of negligible influence.
selected by (1) availability (CS5, EPS) and (2) with A ”weak gold standard” was established using the
respect to their design purpose (THP, KDP), with all 4 monitor ratings of 2 experts obtained in a consensus
belonging to the budget and medium price segment conference. We are well aware that while the test under
ranging between $150 and $650. The very expensive evaluation represents the gold standard, considerable error
CHO printer had been included as a control. Our basic may be introduced.15,16 On the other hand, it is difficult to
establish a ground truth when using patient radiographs.
idea was to create a rather simple test pattern to cali-
Since the monitor evaluation itself is considered the cur-
brate printers in a way that, regardless of their actual
rent gold standard for digital radiograph viewing, how-
printing technique, they were capable of accurately
ever, testing against this standard should provide mean-
printing dental radiographs. Because of the large vari-
ingful results. It should be noted here, that no ”ground
ety of settings for the different printers, our approach
truth” is really representing truth in a universal sense;
was not to specify particular settings but rather specify
rather it represents the best possible information on the
the required printing features, including fundamental
object one can obtain on a state-of-the art level. To avoid
print properties like contrast, spatial resolution, color
bias because of varying image size,17 the monitor display
gradient, and geometric accuracy. A test pattern was
had to approximately match the size of the prints. We
specifically designed to enable printing-quality assur- observed rather high Az values of around 0.8, indicating
ance in the dental office. In the instance of a hardcopy good diagnostic performance as compared to the monitor.
request, to reduce time consumption once a sufficient The clear overlap of the 95% CIs of the mean indicate
setting configuration of the printer is found, it should be nonsignificant differences between them.18 Neither the
recorded and used per default in the subsequent in- high-cost color thermo-sublimation printer (CHO) nor the
stances of printing. Methods to enhance printing quality radiographic image ink-jet printer Tetenal performed bet-
typically include a cleaning procedure of the nozzles ter than their budget competitors. Actually, the signifi-
followed by the adjustments specified earlier. cantly lower subjective ratings (Fig. 4) plus the objec-
We only selected a few radiographs displaying typical tively worse reproduction of test pattern features of the
dental radiographic features such as interproximal carious Tetenal printer seem to be in line with the objective data
lesions, apical translucencies, and the periodontal liga- also located on the low end of the scale. Yet, in light of the
ment space. They had been selected from the patient obvious overlap of the Az 95% CIs, this at most may
records of 2 university hospitals and 4 dental offices. The indicate a trend. Another interesting aspect is that the less
selection process beforehand was applied to (1) provide expensive ink-jet printers Canon S500 and Epson Stylus
typical images with relevant quality and to (2) reduce the Photo were ranked in the upper half in the objective
number of variables. Bearing in mind the low inter-rater evaluation.
reproducibility, variation among several images may have Our findings are in agreement with those of other
clouded the printer effect. In addition, the observer burden authors. Benediktsdottir and Wenzel7 conclude for the
due to the overall number of observations has to be taken assessment of position and morphology of mandibular
into account. For assessment of accuracy, only interprox- third molars, that ink-jet prints on glossy paper ”may be as
imal carious lesions and apical translucencies in the set of accurate as the original monitor-displayed digital pan-
3 radiographs were prompted. Subjective rating of image oramic images.” Another study indicates that paper prints
OOOOE
584 Schulze et al October 2008
of digitized film radiographs compare well to their film- ink jet-printed paper copies: pilot study. J Digit Imaging
based origin with respect to caries detection.8 Guerrant 1994;5:246-51.
4. Lyttkens K, Kirkhorn T, Kehler M, Andersson B, Ebbesen A,
and colleagues5 showed that thermo-sublimation prints Hochbergs P, et al. Evaluation of the image quality of ink-jet
subjectively revealed equal diagnostic utility as monitor printed paper copies of digital chest radiographs as compared
display for various anatomic features. Yet subjective scor- with film: a receiver operating characteristic study. J Digit Im-
ings were found to be significantly better for expensive aging 1994;7:61-8.
direct-thermal film prints when compared to budget ink- 5. Guerrant GH, Moore WS, Murchison DF. Diagnostic utility of
thermal printed panographs compared with corresponding com-
jet prints on glossy paper.6 From a physical point of view, puter monitor images. Gen Dent 2001;49:190-6.
the dynamic transmission range of transparent media is 6. Gijbels F, Sanderink GC, Pauwels H, Jacobs R. Subjective image
inherently larger than the dynamic range of reflection quality of digital panoramic radiographs displayed on monitor
media such as printing paper.19 There are 2 reasons for the and printed on various hardcopy media. Clini Oral Invest
surprisingly good performance of the paper prints. First, 2004;8:25-9.
7. Benediktsdottir IS, Wenzel A. Accuracy of digital panoramic
intrinsic contrast enhancement applied per default by the images displayed on monitor, glossy paper, and film for assess-
manufacturer’s software is a likely factor. Since contrast ment of mandibular third molars. Oral Surg Oral Med Oral
on reflection media is measured quite differently from that Pathol Oral Radiol Endod 2004;98:217-22.
of a transmission medium such as the monitor, however, 8. Otis LL, Sherman RG. Assessing the accuracy of caries diagno-
direct comparison is relatively complicated. Yet the re- sis via radiograph. Film versus print. J Am Dent Assoc
2005;136:330.
sults of the studies indicate that, if contrast enhancement is 9. Funke M, Obenauer S, Hermann KP, Fischer U, Grabbe E.
applied, it seems to work for radiographic evaluation also. Softcopy- versus Hardcopybefundung in der digitalen Mammog-
Second, for prints acquired in color mode, because of the raphie. Radiologe 2002;42:265-9.
superior color perception of humans the contrast is artifi- 10. Ishigaki T, Endo T, Ikeda M, Kono M, Yoshida S, Ikezoe J,
cially enhanced. Unfortunately, not all manufacturers al- Murata K, Matsumoto M. Subtle pulmonary disease: detection
with computed radiography versus conventional chest radiogra-
low the user to control for the printing mode (color or
phy. Radiology 2005;201:51-60.
black and white). Consequently, we set aside this issue 11. Kosuda S, Kaji T, Kobayashi H, Watanabe M, Iwasaki Y, Ku-
according to our basic approach, that no matter what the sano S. Hard-copy versus soft-copy with and without simple
actual settings may be, as long as the test pattern features image manipulation for detection of pulmonary modules and
are reproduced adequately, the print quality should be masses. Acta Radiol 2000;41:420-4.
12. Goldmann M, Pearson AH, Darzenta N. Endodontic success—
sufficient. Obviously, our preliminary findings plus those
Who’s reading the radiograph? Oral Surg 1972;33:432-7.
from other authors5,7,8 support this assumption. Because 13. Pascoal A, Lawinski CP, Honey I, Blake P. Evaluation of a
some test features required a priori were not reproduced software package for automated quality assessment of contrast
sufficiently by all printers, we conclude that the pattern detail images— comparison with subjective visual assessment.
introduced here may be even too demanding. Further Phys Med Biol 2005;50:5743-57.
14. Wenzel A. Digital radiography and caries diagnosis. Dentomax-
work has to be done in reducing the test features to the
illofac Radiol 1998;27:3-11.
necessary minimum. 15. Wenzel A, Hintze H. Comparison of microscopy and radiogra-
phy as gold standards in radiographic caries diagnosis. Den-
CONCLUSION tomaxillofac Radiol 1999;28:182-5.
The results of this preliminary study are promising 16. Wenzel A, Hintze H. Editorial review. The choice of gold stan-
dard for evaluating tests for caries diagnosis. Dentomaxillofac
regarding the diagnostic accuracy on glossy-paper Radiol 2001;1999:132-6.
prints obtained from budget printers for typical dental 17. Versteeg KH, Sanderink GCH, Geraets WGM, van der Stelt PF.
radiographic diagnoses. Because of the preliminary na- Impact of scale standardization on images of digital radiography
ture of this study and the rather thin databases available systems. Dentomaxillofac Radiol 1997;26:337-43.
in the literature, however, further studies are required 18. Henderson RA. Assessing test accuracy and its clinical conse-
quences: a primer for receiver operating characteristic curve
before a final statement about the safety of such low- analysis. Ann Clin Biochem 1993;30:521-39.
cost hardcopies in clinical use is feasible. 19. Weinstein LM, Fitzer MS, Fitzer PM. Detail enhancement in
prints of radiographs. Use of a linear radial transmission filter.
REFERENCES Radiology 1975;115:726-8.
1. Bundesregierung BRD. Verordnung zur Änderung der Röntgen-
verordnung und anderer atomrechtlicher Verordnungen. Bundes-
druckerei GmbH, Berlin, Germany, 2002. APPENDIX 1. (QUESTIONNAIRE)
2. Richtlinie zur Durchführung der Qualitätssicherung bei Röntge- Dear observer,
neinrichtungen zur Untersuchung oder Behandlung von Men- You will be presented with 3 intraoral radiographs
schen - Qualitätssicherungs-Richtlinie (QS-RL). Gemeinsames (No. 1, 2, 3). Please, position them separately on the
Ministerialblatt, Carl Heymanns Verlag KG, Cologne, Germany,
2004.
white cardboard enclosed in such a way that the number
3. Kirkhorn T, Kehler M, Nilsson J, Lyttkens K, Andersson B, is clearly readable in the left upper corner of the
Holmer N-G. Demonstration of digital radiographs by means of image. The radiographs should be evaluated one after
OOOOE
Volume 106, Number 4 Schulze et al 585
another using an indirect lighting environment. Please tooth 47 □ definitely caries present
answer the questions in the questionnaire for each sin- mesially □ uncertain
gle radiograph according to their numbering, and check □ definitely no caries present
the answer best matching your decision for each 1.3. How would you judge the radiographic diagnos-
question (check only 1 answer per question!). tic significance of the marginal bone level separately
Please note: If questions refer to structures (teeth, for each interproximal region?
etc.) only partly visible on the radiographs, the ques- 1.3.1
tions are only applicable to that particular part of between tooth 44 □ diagnostically acceptable
and □ uncertain
the structure. tooth 45 □ diagnostically not acceptable
1.3.2
RADIOGRAPH NO. 1 between tooth 45 □ diagnostically acceptable
From the left to the right in this radiograph teeth no. and □ uncertain
tooth 46 □ diagnostically not acceptable
47, 46, 45, and 44 are (partly) visible.
1.3.3
1.1. How do you judge the radiographic diagnostic
between tooth 46 □ diagnostically acceptable
significance of the periodontal ligament gap for each and □ uncertain
tooth? tooth 47 □ diagnostically not acceptable
1.1.1 1.4. How would you judge the radiographic diagnos-
tooth 47 □ diagnostically acceptable tic significance of the filling in tooth 46 in the distal
□ uncertain approximal region?
□ diagnostically not acceptable
□ there is certainly a gap between filling and tooth
1.1.2 □ uncertain
tooth 46 □ diagnostically acceptable □ there is certainly no gap between filling and tooth
□ uncertain □ no filling is visible in tooth 46
□ diagnostically not acceptable
1.1.3
tooth 45 □ diagnostically acceptable
□ uncertain
RADIOGRAPH NO. 2
□ diagnostically not acceptable From the left to the right you (partly) see teeth no.
1.1.4 24, 25, 26, and 27.
tooth 44 □ diagnostically acceptable 2.1. How would you judge the diagnostic signifi-
□ uncertain cance of the radiographic periodontal ligament gap
□ diagnostically not acceptable separately for each tooth?
1.2. Please indicate your decision about the presence 2.1.1
or absence of an interproximal caries lesion separately tooth 24 □ diagnostically acceptable
for each interproximal region. □ uncertain
1.2.1 □ diagnostically not acceptable
tooth 44 □ definitely caries present 2.1.2
distally □ uncertain tooth 25 □ diagnostically acceptable
□ definitely no caries present □ uncertain
1.2.2 □ diagnostically not acceptable
tooth 45 □ definitely caries present
2.1.3
mesially □ uncertain tooth 26 □ diagnostically acceptable
□ definitely no caries present □ uncertain
1.2.3 □ diagnostically not acceptable
tooth 45 □ definitely caries present
2.1.4
distally □ uncertain tooth 27 □ diagnostically acceptable
□ definitely no caries present □ uncertain
1.2.4 □ diagnostically not acceptable
tooth 46 □ definitely caries present
2.2. How would you judge the radiographic diagnos-
mesially □ uncertain tic significance of the marginal bone level separately
□ definitely no caries present for each single interproximal region?
1.2.5 2.2.1
tooth 46 □ definitely caries present between tooth 24 □ diagnostically acceptable
distally □ uncertain and □ uncertain
□ definitely no caries present tooth 25 □ diagnostically not acceptable
1.2.6 2.2.2
OOOOE
586 Schulze et al October 2008