Factors determining the diagnostic

accuracy of digitized conventional

intraoral radiographs
M. Obki, T. Okano * and T. Nakamura
Department of Oral Radiology, Nagasaki University School of Dentistry, Nagasaki and *Department of Oral
Radiology, Showa University School of Dentistry, Tokyo, Japan

Received 6 May 1993, and in final form 12 July 1993

A number of factors may determine the diagnostic accuracy of digitized intraoral radiographs.
Conventional film radiographs were digitized with three different digitizers, a laser-scanner, a
drum-scanner and a TV camera. Digital images, varying in pixel size, grey level and image
processing, were displayed on three different video-monitors and assessed by 10 dentists. The
detectability of incipient proximal surface caries was used as an index of the diagnostic
accuracy and the effect of the different factors compared by means of receiver operating
characteristic (ROC) analysis. Images digitized by the drum-scanner were found to have the
best diagnostic accuracy. Sufficient diagnostic accuracy could be attained on the low-cost video
monitor of a personal computer. Digital images with a pixel size of 100}Lm and 32 grey levels
were found to be acceptable for intraoral radiographs for diagnostic purposes. These results
provide a practical guide for establishing a digital image acquiring system for any intraoral
radiographs, reducing demands on data storage to a minimum.
Keywords: Radiographic image enhancement; radiography, dental; ROC analysis; dental caries

Dentomaxillofac. Radiol., 1994, Vol. 23, 77-82, May

Digital imaging has the potential for improving
diagnostic accuracy and making quantitative diagnoses.
In addition, it makes a picture archiving and communi-
cation system (PACS) feasible. However, it is neces-
sary in the first instance to establish that the accuracy of
digital images is sufficient for a specific diagnostic
purpose. The diagnostic accuracy of digital chest
radiographs displayed in various formats has been
investigated':", and in dental radiographs, digitization
has been evaluated in terms of the detectability of
proximal and occlusal caries 7- 13 . Kassebaum et al. 12
assessed the appropriate spatial resolution for a
commercially available digital imaging and transmis-
sion system and could find no significant differences
between the digital images with pixel sizes of 200, 300
or 4OO}Lm, or between these and the original film
images. Dove and McDavid 13 used a TV-based image
processing system but could not detect any difference
between non-enhanced digital images and conventional
film-based images 13. The histogram equalization
technique was applied to the digital images but did not
improve diagnostic accuracy 13 • A number of studies
have failed to show any clear advantage of computer-
based image processing and analysis over conventional
radiographs for improving the detectability of proximal
cariesl 4-17 . These previous studies used only a
TV-based digital imaging system, but the physical
characteristics of the digitization devices and viewing
monitors may also have a profound effect on diagnostic
© 1994Butterworth-Heinemann Ltd for the IADMFR
025G-832X/94/020077-06
accuracy. The digitization process is especially impor-
tant so as not to lose small changes in radiographic
density such as the subtle radiolucencies caused by
incipient proximal caries. If such subtle information is
decreased or lost in the digitization process, then it is
difficult to see how the digitized image will produce a
higher level of diagnostic accuracy than the original
radiograph, even if the image processing is performed
after digitization. Recently, direct digital imaging in the
form of RadioVisioGraphyl 8-20 has been introduced
and its diagnostic accuracy reported. Although promis-
ing, some problems still have to be resolved before such
methods can be substituted for conventional radio-
graphy in dental practice. A significant problem may be
the smaller optical density range and greater image
noise in digitized images when compared with radio-
graphs. Therefore, for the present, digitization of film
radiographs is still important for the development of
effective image processing procedures for intraoral
radiography. It is also necessary if a PACS is to be
implemented for the accumulated radiographs in dental
facilities so as to perform retrospective image analysis
on them.
In this study we have performed a further investiga-
tion into the diagnostic accuracy of digitized intraoral
radiographs to clarify the effects of digitization proce-
dure, viewing monitors and image processing compar-
ing three types of digitizer and three video-monitors for
this purpose.
Dentomaxillofac. Radiol., 1994, Vol. 23, May 77
Diagnostic accuracy of digitized intraoral radiographs: M. Ohki et al.

Materials and methods
X-ray film digitizers
An X-ray film digitizer converts film densities into
numeric grey levels for input into a digital computer.
The most commonly used digitizers are microdensito-
meters, laser film scanners and TV cameras. In micro-
densitometers, the X-ray film is mounted on a flat bed
or wrapped around a drum and a beam of light scans
the film as the bed translates or the drum rotates for
two-dimensional imaging. In laser film scanners the
film is translated mechanically and scanned by a narrow
laser beam. In both devices the intensity of the
transmitted light is read by a photomultiplier and
converted to film densities. With TV cameras, the film
is placed on a light box which is focused directly onto a
photosensitive tube or a charge-coupled device (CCD)
sensor. These photodetectors have a smaller dynamic
range than a photomultiplier. The intensity of the
transmitted light is not converted into film density and
depends on the brightness of the light box used with the
camera.
Three types of digitizers were compared; a drum
scanning microdensitometer (Model 2605, Abe-sekkei
Co., Tokyo, Japan), a laser film scanner (Model 2903,
Abe-sekkei Co. Tokyo, Japan) and a TV camera
(Model C1000, Hamamatsu Photo. Co., Shizuoka,
Japan). Both the drum-scanner and the laser-scanner
recorded film densities ranging from 0 to 4 and the
output signals were 8-bit and lO-bit data which corres-
pond to 256 and 1024 grey levels, respectively. Spatial
resolution varied with the size of the scanning beam
and of the sampling aperture. Intraoral radiographs
were digitized with pixel sizes of 50 JLm, the minimum
for our laser-scanner, and 100 JLm. The lO-bit output
data from the laser-scanner were linearly converted
into 8-bit data (256 grey levels). For the TV camera,
the magnification was set so that one pixel corres-
ponded to 50 JLm, and the output signal converted to
8-bit data.
Viewing monitors
The digital images were displayed on three video
monitors: a 13-inch (33-cm) Apple RGB colour monitor
for a Macintosh personal computer (Apple Computer,
Inc., Cupertino, CA, USA), a 17-inch (43-cm) monitor
for a VIEW2000 imaging workstation (Virtual Imag-
ing, Sunnyvale, CA, USA) and a 16-inch (41-cm)
IMLOGIX image viewing system (Genesys/Imlogix,
Fenton, MO, USA). Although it is not designed
specifically for viewing X-ray images, the Apple RGB
monitor can display 24-bit colour images and 8-bit
monochrome images in 640 X 480 pixels. The
VIEW2000 monitor displays 16-bit monochrome
images in 780 X 580 pixels. The IMLOGIX displays
images up to 1024 X 1024 pixels in 12 bits. Both are
designed for viewing X-ray images and can function
as part of PACS. Magnification, window level and
window width can be easily adjusted by the viewer
using a mouse.
Radiographs and digital images
The intraoral radiographs used in this study were
taken from our previous study". Extracted human
premolars, with or without discolouration on the
proximal surfaces, were radiographed using a geometry
consistent with bitewing projections. The focus-object
distance was 40 em, and the object-film distance
0.5 em. The dental X-ray unit (Model R802, Tokyo
Emix Co., Tokyo, Japan) was operated at 65 kVp with
2.0 mm total Al equivalent filtration. Exposure time
was adjusted so as to achieve an optical density of the
proximal enamel of 0.6-0.8. Ultraspeed (Eastman-
Kodak, Rochester, NY, USA) film was used and
developed for 5 min at 20°C. Following radiography,
each tooth was embedded in epoxy resin and mesio-
distal ground sections obtained to confirm the
presence, if any, and extent of caries. A total of 64
proximal surfaces were used in this study. Thirty
surfaces were caries-free, 11 had caries limited to the
outer half of the enamel and 13 to the inner half, and in
10 caries had penetrated into dentine.
All the radiographs were digitized using one of the 10
modalities shown in Table I. The effect of varying bit
depth, pixel size and image processing on diagnostic
accuracy was evaluated on the images digitized by

Table I The effects of the digitizer, pixel size, grey level, image processing and monitor on diagnostic accuracy, expressed as the area under the
ROC curve, P(A). Values are the mean (± s.d.) ofthe 10 observers. Probabilities were obtained by two-tailed paired r-tests with the original film
with a significance level set at 95% '

Pixel Grey
Images Digitizer size level Image processing Monitor P(A) s.d. Probability Significance

No.1 Drum 50JLm 256 None A 0.722 0.053 0.0147 S

B 0.747 0.053 0.1865 NS
C 0.764 0.068 0.4666 NS
C' 0.691 0.070 0.0162 S
No.2 Drum 50JLm 256 Edge enhanced A 0.752 0.044 0.3052 NS
No.3 Drum 50JLm 32 Gray level reduced A 0.725 0.049 0.0342 S
No.4 Drum 50JLm 64 Gray level reduced A 0.715 0.063 0.0052 S
No.5 Drum looJLm 256 x2 magnified A 0.700 0.062 0.0278 S
No.6 Drum looJLm 256 x2 magnified with interpolation A 0.681 0.070 0.0021 S
No.7 Laser 50JLm 256 None A 0.675 0.049 0.0002 S
No.8 Laser 50JLm 256 Contrast enhanced A 0.679 0.047 0.0014 S
No.9 Laser 50JLm 256 Contrast and edge enhanced A 0.692 0.065 0.0018 S
No. 10 TV 50JLm 256 None A 0.679 0.061 0.0010 S
Original 0.781 0.061
film
S. significant; NS. not significant.
Monitors: A. Apple RGB; B. VIEW2000; C. IMLOGlX; C·. IMLOGIX with user interaction.

78 Dentomaxillofac. Radiol., 1994, Vol. 23, May

 Diagnostic accuracy of digitized intraoral radiographs: M. Ohki et al.

the drum-scanner and displayed on the Apple RGB
monitor. Images with 32 and 64 grey levels were
reproduced from those with pixel sizes of 50 JLm and
256grey levels by reducing the bit depth. Images with a
pixel size of 100JLm were magnified twice so as to be
the same size as those with a pixel size of 50 JLm on
the video-monitor. The bilinear interpolation for
smoothing was also applied to the magnified images.
Edge enhancement was performed with a high-pass
mask after median filtering for noise reductiorr'",
Contrast enhancement was performed in converting the
10-bit output data from a laser-scanner into 8-bit
data, i.e. the grey levels between the minimum and
maximum levels on the image digitized by a laser-
scanner were linearly scaled to the range between 0 and
255. The three types of viewing monitors were com-
pared with each other for images digitized with a pixel
size of 50 JLm and 256 grey levels. The effect of the
observers' manipulation of the viewing conditions on
the IMLOGIX monitor was also evaluated.
ROC analysis
Ten dentists of whom six were oral radiologists, two
preventive dentists and two periodontists served as
observers. All had been engaged in clinical practice for
255.------------------...,
192
a;
> intraoral radiograph of a tooth where caries had
~ 128
~ proximal surfaces. The images shown in Figure 2 were
CJ
64
o on the way in which the digitizer recorded the density.
o
0.0 1.0 2.0 3.0
Optical film density
Figure 1 Comparison of the efficacy of the three digitizers
( • , drum-scanner; ---0--, laser-scanner; - . A - ,
TV camera) in converting optical density to grey level
more than 3 years. All the original radiographs and
digitized images, from the total 64 tooth surfaces, were
presented to them at random. They were asked to
decide whether or not caries was present and to rate
their confidence on a five-point scale: (1) definitely
present, (2) probably present, (3) unsure, (4) probably
absent, (5) definitely absent. All viewing was per-
formed under dim indoor lighting. The brightness and
the contrast level of the monitor were fixed, except for
the case of the IMLOGIX monitor, where observers
were allowed to change the window level, window
width, and zoom factor. No time limit was given for
any viewing sessions. The diagnostic accuracy was
determined as the area under the ROC curve, P(A),
which was obtained by integrating multiple trapezoidal
areas of the connected ROC data points-", A paired
r-test and one-factor ANOVA were used for statistical
analysis.
Results
Figure 1 compares the ability of the three digitizers
to record the film density in terms of the grey level.
Both the laser-scanner and the drum-scanner had a
very accurate linear relationship. In contrast, the TV
camera had a narrow dynamic range with an upper limit
of readable film density of about 2. The relationship
was not linear, nor was it possible to read film density
directly, because the value depended on the brightness
of the light box and on the size of the aperture of the
camera.
Figures 2 and 3 show typical digital images of an
penetrated to the inner half of the enamel on both
obtained with each of the digitizers with the same pixel
size of 50 JLm and 256 grey levels. Although all three
images were displayed in the same grey level region
ranging from 0 to 255, the contrast differed, depending
4.0 The outline of the tooth in images digitized by the laser-
scanner (Figure 2b) appeared slightly ill-defined,
whereas those digitized by the TV camera showed the
similar contrast to the original radiograph when viewed
on a light-box. Figure 3 shows an example of an image

Figure 2 Typical images obtained at 50JLm pixel size and 256 grey levels following digitization by the three different digitizers. a, By a drum-
scanner; b, by a laser-scanner; c, by a TV camera. Caries has extended on both proximal surfaces into the inner-half of enamel

Dentomaxillofac. Radio!., 1994, Vo!. 23, May 79

Diagnostic accuracy of digitized intraoral radiographs: M. Ohki et al.

Figure 3 Digital images of the same tooth as in Figure 2 showing the effects of varying pixel size, grey level and image processing. a, 50 ILm pixel
size and 32 grey levels; b, lOOlLm pixel size and 256 grey levels; c, 50ILm pixel size and 256 grey levels with edge enhancement

digitized in various modalities by the drum-scanner. 1.0

The image with a pixel size of 50porn and 32 grey levels g
a.
had a little grey level contouring which would be 0>
acceptable. The image with a pixel size of 100JLm and ~

I I I I
:::J
0
256 grey levels which was magnified twice had a barely
8a:
noticeable checker-board effect. This effect could be
reduced by an interpolation at magnification. Figure 3c
shows the effect of edge enhancement on an image with
a pixel size of 50porn and 256 grey levels.
lii
'C
c:
:::J
l'll
~
-e 0.5
I
Table I compares P(A) values of all the digital L - _ - - l . ._ _....l.-_ _.L..-_--l.._ _- I . - _ - - - I
images with that of the original radiographs. A Original monitor A monitor B monaor C monitor C
significant difference was found with the Apple film with user interadion
monitor and with the IMLOGIX with user inter-
Figure 5 Comparison of the effect of the viewing monitors on
action when the images were digitized by the drum- diagnostic accuracy of digital images on various viewing monitors
scanner with a pixel size of 50porn and 256 grey levels. (expressed as the area under the ROC curve P(A». Images were
In contrast, the images with edge enhancement on the displayed with 50ILm pixel size and 256 gray levels following
Apple monitor did not show a significant difference. digitization by a drum-scanner. Monitor A, Apple RGB monitor;
Monitor B, VIEW2000; Monitor C, IMLOGIX. Scale bar
Figures 4-7 analyse the effects that the various equals ± 1 s.d.
factors had on diagnostic accuracy. Figure 4 compares
the three types of digitizers. All the images had a pixel
size of 50 JLm and 256 grey levels and were viewed on was no significant difference between the three
an Apple RGB monitor (Table I, images lA, 7 and digitizers.
10). Paired r-test indicated statistically significant Figure 5 analyses the effect of the three types of
differences between the original film and the digital monitors on diagnostic accuracy. The images were
images. Although the diagnostic accuracy was highest digitized by the drum-scanner with a pixel size of
with the drum-scanner, ANOVA showed that there 50 JLm and 256 grey levels. There was no significant
difference between the images on any of the
1.0 monitors by ANOVA. The IMLOGIX monitor had the
g highest diagnostic accuracy of all three. However,
a.. when observers were allowed to change window level,
0;
e:::J window width and magnification of the images on this
U

s I I monitor, there was a significant decrease in diagnostic

accuracy.

I
II:
lii
'C
c:
:::J
l'll
I Figure 6 examines the effect on diagnostic accuracy
of grey levels. Images were digitized with a pixel size of
50 JLm at 32, 64 and 256 grey levels and viewed on the
e
-c
Apple RGB monitor. ANOVA showed there was no
0.5 significant difference between the images at any grey
Original Drum- Laser- TV levels.
film scanner scanner camera Figure 7 shows the effect on diagnostic accuracy of
pixel size and image processing. All the images were
Figure 4 Comparison of diagnostic accuracy, expressed as the area digitized with 256 grey levels and viewed on the Apple
under the ROC curve P(A), of intraoral radiographs with images
digitized by the three digitizers. All digital images had 50 JLm pixel RGB monitor. The paired r-test indicated that there
size and 256 grey levels and were displayed on an Apple RGB was no significant difference between the original film
monitor. Scale bar equals ± 1s.d. and the 50porn images treated with edge enhancement.

80 Dentomaxillofac. Radiol., 1994, Vol. 23, May

 Diagnostic accuracy of digitized intraoral radiographs: M. Ohki et al.

1.0 of 0.888 for bitewing radiographs calculated by the

«Ci:' trapezoidal method. This value is higher than those
ti
obtained by both Dove et 01. 13 and us. One reason may
~ be that the caries was deeper in the study of Kassebaum

I ! I I
:::J
U
et alY. These authors concluded that there was no
8a: significant difference in the diagnostic accuracy
~
between images digitized by a TV camera with a pixel
'C
c::: size of 300 p.m and the original radiographs. This result
:::J
III also seems to be affected by their case samples. We
l!? found that images with a pixel size of more than
«
0.5 100p.m had a noticeable checker-board effect which
Original 32 64 256graylevels clearly lowered diagnostic accuracy. With a screen-film
film combination system, a pixel size of 4oop.m has been
Figure 6 Comparison of. the effect of varying grey level on the reported to be sufficient for digitization". On the other
diagnostic accuracy of digital images expressed as the area under the hand, the direct film system used in intraoral radio-
ROC curve P.(A). Images were digitized with 50ILm pixel size by a
drum-scanner and displayed on an Apple RGB monitor. Scale bar graphy has a higher spatial resolution of the order of
equals ± 1 s.d. 20 p.m. Therefore, a digital image with a pixel size of
more than 100p.m will probably lose diagnostic infor-
1.0 mation. Nonetheless, no significant difference in the
sa.. diagnostic accuracy could be found between images
digitized with pixel sizes of 50 p.m and 100 p.m. The
Gi
~ pixel size of 100 p.m has the added advantage of

I I
:::J
U reducing the amount of data storage required to one-
8a: I fourth of that with a pixel size of 50 p.m. We also
...CD
'C
c:::
:::J
III
I I
l!?
-e
0.5
Original
SOllm*
film
Pixel size
Figure 7 Comparison of the effect of varying pixel size on diagnostic
accuracy of digital images expressed as the area under the ROC curve
P(A). Images were digitized with 256 grey levels by a drum-scanner
and displayed on an Apple RGB monitor. Edge enhancement was
performed on the 50lLm* images, and the interpolation was used on
the x 2 magnified lOOlLm* images
No significant difference could be detected between the
images with pixel sizes of 50 p.m and 100p.m. Interpo-
lation for reducing dither patterns in the 2 x magnified
images did not improve diagnostic accuracy. The effect
of image processing on diagnostic accuracy of the
images digitized by the laser-scanner is shown in Table
1, images 7,8 and 9. Although edge enhanced images
showed a slightly higher diagnostic accuracy, the
difference was not significant.
Discussion
We calculated the area under ROC curves as the sum of
the trapezoidal area P(A) because spread of data points
was such that the curve fitting method, by which the Az
value is derived, could not be applied successfully.
Estimating P(A) by this method gives somewhat lower
values than for Az, but this means that the diagnostic
accuracy is estimated more strictly. Dove et 01. 8 found
the diagnostic accuracy of intraoral radiographs for
proximal caries was 0.7586; this value was obtained
solely on the basis of the presence of caries, regardless
of the lesion depth. Their values for digital images
calculated as Az values by the curve fit method ranged
from 0.666 to 0.7173. These figures compare favour-
ably with our findings of 0.781 for the original film and
0.679 for the images digitized by a TV camera.
Kassebaum et 01. 12 , on the other hand, reported a value
confirmed that grey level can be reduced to 32 levels
without loss of diagnostic accuracy. This must be due to
the fact that the human eye can distinguish at most
about 50 contrast levels. An image with 32 grey levels
requires one-eighth the data storage of an image with
256 grey levels. 16 grey levels (4-bit image) are insuffi-
cient since, in this investigation, they produced a
noticeable grey level contouring.
The drum-scanner showed a higher diagnostic
accuracy than the laser-scanner and the TV camera.
Although the laser-scanner was as accurate for reading
film density as the drum-scanner, the lower diagnostic
accuracy was probably due to the fact that it produced
an ill-defined outline of the tooth. This effect might be
caused by the non-uniformity of the polygon mirror in
the laser-scanner which reflects the laser beam for
scanning. Overall, no significant difference was
detected between the TV camera and the other more
accurate digitizers. Although low-cost TV cameras are
available for constructing a digital image acquiring
system, the lower dynamic range may restrict their
potential for improving diagnostic accuracy by an
additional image processing.
The installation of a large number of monitors is
required for a PACS. The cost is therefore an
important consideration. It is apparent from our study
that a video-monitor designed for a personal computer
is acceptable for this purpose. We also found that
window operations such as changing brightness, con-
trast and magnification did not, in fact, improve
diagnostic accuracy but, on the contrary, often reduced
it. This may be because the digital images had already
been created in the optimal density region. The
histogram equalization technique which Dove et 01. 13
evaluated also decreased diagnostic accuracy. In this
investigation, edge enhancement improved diagnostic
accuracy. We believe it is therefore possible to produce
an image with higher diagnostic accuracy than that of
the original radiograph by developing a suitable image
processing technique. An accurate digitization device
such as a drum-scanner may be needed for this
purpose.

Dentomaxillofac. Radiol., 1994, Vol. 23, May 81

Diagnostic accuracy of digitized intraoral radiographs: M. Ohki et al.
In conclusion, we have confirmed that highly
accurate digitization devices and viewing monitors are
not necessary if digital images are used only for
diagnostic purposes in place of conventional intraoral
radiographs. We also found that digitization with a
pixel size of 100um and 32 grey levels for intraoral
radiographs was similarly acceptable.
References
1. Carterette EC, Fiske RA, Huang HK. Receiver operating
characteristic (ROC) evaluation of a digital viewing station for
radiologists. SPIE Proc 1986; 626: 441-6.
2. MacMahon H, Metz CE, Doi K, Kim T, Giger ML, Chan HP.
Digital chest radiography: effect on diagnostic accuracy of hard
copy, conventional video, and reversed gray scale video display
formats. Radiology 1988; 168: 669-73.
3. Hayrapetian A, Aberle D, Huang HK, Fiske R, Morioka C,
Valentino D, Boechat MI. Comparison of 2048-line digital
display formats and conventional radiographs: an ROC study.
AIR 1989; 152: 1113-8.
4. Lams PM, Cocklin ML. Spatial resolution requirements for
digital chest radiographs: an ROC study of observer performance
in selected cases. Radiology 1986; 158: 11-9.
5. MacMahon H, Vyborny CJ, Metz CE, Doi K, Sabeti V,
Solomon SL. Digital radiography of subtle pulmonary abnorma-
lities: an ROC study of the pixel size on observer performance.
Radiology 1986; 158: 21-6.
6. Goodman LR, Foley WO, Wilson CR, Rimm AA, Lawson TL.
Digital conventional chest images: observer performance with
film digital radiography system. Radiology 1986; 158: 27-33.
7. Wenzel A, Fejerskov 0, Kidd E, Joyston-Bechal S, Groeneveld
A. Depth of occlusal caries assessed clinically, by conventional
film radiographs, and by digitized, processed radiographs. Caries
Res 1990; 24: 327-33.
8. Wenzel A, Hintze H, Mikkelsen L, Mouyen F. Radiographic
detection of occlusal caries in noncavitated teeth. Oral Surg Oral
Med Oral Patholl99l; 72: 621-6.
9. Wenzel A, Larsen MJ, Fejerskov O. Detection of occlusal caries
without cavitation by visual inspection, film radiographs, xero-
radiographs, and digitized radiographs. Caries Res 1991; 25:
365-71.
10. Wenzel A, Fejerskov O. Validity of diagnosis of questionable
caries lesions in occlusal surfaces of extracted third molars.
Caries Res 1992; 26: 188-94.
82 DentomaxiHofac. Radiol., 1994, Vol. 23, May
11. Wenzel A, Verdonschot EH, Truin GJ, Konig KG. Accuracy of
visual inspection, fiber-optic transillumination, and various
radiographic image modalities for the detection of occlusal caries
in extracted noncavitated teeth. I Dent Res 1992; 71: 1934-7.
12. Kassebaum DK, MacDavid WO, Dove SB, Waggener RG.
Spatial resolution requirements for digitizing dental radiographs.
Oral Surg Oral Med Oral Patho11989; 67: 760-9.
13. Dove SB, McDavid WO. A comparison of conventional intra-
oral radiography and computer imaging techniques for the
detection of proximal surface dental caries. Dentomaxillofac
Radioll992; 21: 127-34.
14. Pitts NB. Detection and measurement of approximal radio-
lucencies by computer-aided image analysis. Oral Surg Oral Med
Oral PathoI1984; 58: 358-66.
15. Pitts NB, Renson CE. Further development of a computer-aided
image analysis method of quantifying radiolucencies in approxi-
mal enamel. Caries Res 1986; 20: 361-70.
16. Heaven TJ, Firestone AR, Feagin FF. Computer-based image
analysis of natural approximal caries on radiographic films. I
Dent Res 1992; 71: 846-9.
17. Verdonschot EH, Kuijpers JMC, Polder BJ, De Leng-Worm
MH, Bronkhorst EM. Effects of digital grey-scale modification
on the diagnosis of small approximal caries lesions. I Dent 1992;
20: 44-9.
18. Benz C, Mouyen F. Evaluation of the new RadioVisioGraphy
system image quality. Oral Surg Oral Med Oral Patholl991; 72:
627-31.
19. Wenzel A, Hintze H, Mikkelsen L, Mouyen F. Radiographic
detection of occlusal caries in noncavitated teeth. A comparison
of conventional film radiographs, digitized film radiographs, and
RadioVisioGraphy. Oral Surg Oral Med Oral Patholl991; 72:
621-6.
20. Russell M, Pitts NB. Radiovisiographic diagnosis of dental
caries: initial comparison of basic mode videoprints with bitewing
radiography. Caries Res 1993; 27: 65-70.
21. Okano T, Ohki M, Huang HJ, Yamada N. Diagnosis of non-
cavernous posterior proximal caries: radiographic observer
performance. Oral Radio11987; 3: 1-6.
22. Pratt WK. Digital image processing. New York: Wiley-
Interscience, 1978.
23. Swets JA, Pickett RM. Evaluation ofdiagnostic systems: methods
from signal detection theory. New York: Academic, 1982.
Address: Dr Masafumi Ohki, Department of Oral Radiology, Naga-
saki University School of Dentistry, 1-7-1 Sakamoto, Nagasaki
852, Japan.