Dentomaxillofacial Radiology (1999) 28, 152 ± 157

ã 1999 Stockton Press All rights reserved 0250 ± 832X/99 $12.00

http://www.stockton-press.co.uk/dmfr

Two- and three-dimensional imaging modalities for the detection of

caries. A comparison between ®lm, digital radiography and tuned
aperture computed tomography (TACTTM)
M Abreu Jr*,1,2, D A Tyndall2, E Platin2, J B Ludlow2 and C Phillips3
1
Division of Oral and Maxillofacial Radiology, Federal University of Santa Catarina School of Dentistry, Florianopolis, SC, Brasil;
2
Division of Oral and Maxillofacial Radiology, University of North Carolina School of Dentistry, Chapel Hill, NC, USA and
3
Department of Orthondontics, University of North Carolina School of Dentistry, Chapel Hill, NC, USA.

Objectives: To compare the diagnostic performance provided by two- (®lm and digital
radiography) and three-dimensional imaging modalities (TACT slices and TACT pseudoholo-
grams) in the detection of primary caries.
Methods: Forty-two extracted human posterior teeth were mounted and imaged with
conventional ®lm and direct digital radiography. Free-hand positioning of a dental X-ray
source was used for all exposures. From the digital images, iteratively restored TACT slices and
TACT pseudoholograms were generated. Film images were viewed on a viewbox. Digital
format images were viewed on a high-resolution monitor. Eight observers used a ®ve-point scale
to score the presence or absence of occlusal and proximal caries using the four image
modalities. Observers' assessments were compared with the histological examination of tooth
sections. Possible di€erences in ROC curve areas among image modalities, observers, and
surfaces were assessed by ANOVA. Intra- and interobserver reliability as indicated by intraclass
correlation was also calculated.
Results: There were no statistically signi®cant di€erences between the diagnostic performances
of ®lm, digital radiography, TACT slices and TACT pseudoholograms in the detection of caries
(P=0.310). Intraclass correlation indicated the highest concordance both within and between
observers when ®lm was used for the evaluation.
Conclusions: Under the experimental conditions of this study, three-dimensional TACT
images did not improve caries detection over ®lm or digital radiography. Further research
should investigate the e€ects of imaging variables on TACT's diagnostic ecacy.

Keywords: radiography, dental; digital radiography, dental; dental caries; ROC curve

Introduction

Conventional intra-oral ®lm continues to be the most
widely used radiographic modality for the diagnosis of
dental caries. However, its diagnostic performance is
not entirely satisfactory. Various studies have reported
sensitivity values ranging from 0.40 ± 0.60.1,2 The
introduction of direct digital radiography (DR) has
not helped to solve the problem. In a recent review,
Wenzel reported that with the exception of one study
that showed signi®cantly higher accuracy for DR in
*Correspondence to: M Abreu, Oral & Maxillofacial Radiology, University of
North Carolina School of Dentistry, Chapel Hill, North Carolina, 27599-7450,
USA
Received 15 September 1998; accepted 18 January 1999
comparison with ®lm,3 most studies report similar
performance between these two image modalities.4
Tuned aperture computed tomography (TACTTM), a
more generalized application of the principles of
tomosynthesis,5,6 may help to improve accuracy in
caries diagnosis because of its three-dimensional (3-D),
or pseudo-three-dimensional, capabilities. A complete
review of TACT's theory has been published pre-
viously.7 TACT reconstruction provides the ability to
sample slices of anatomy at varying levels within a
structure. Similarly, pseudo-tridimensional representa-
tions of an object, known as pseudoholograms, can be
generated by sequentially displaying two-dimensional
image frames from di€erent angles. This simulates
varying projection geometries and provides some

perception of three dimensions to the viewer. Studies
have demonstrated TACT's usefulness as a diagnostic
tool.8 ± 11 A previous study of primary caries detection
showed no signi®cant di€erences between the diagnos-
tic accuracy of TACT and ®lm.12 However, the digital
detector that was used provided limited image contrast
due to a reduced dynamic range and only the most
central TACT slice of the teeth was viewed by the
observers. The study also used a specially designed
eight-tube X-ray source that is not commercially
available and therefore does not represent existing
equipment in a typical dental practice.
The aim of this study was to further evaluate TACT
3-D features in the diagnosis of dental caries, by
comparing two kinds of 2-D image modalities,
conventional ®lm and DR, with two kinds of 3-D
image modalities, iteratively restored TACT slices and
TACT pseudoholograms, under conditions that simu-
late real clinical applications. Our null hypothesis was
that there is no signi®cant di€erence in the diagnostic
performance of these four image modalities in assessing
caries.
Materials and methods
Sample preparation and image acquisition
Forty-two extracted human posterior teeth (21 pre-
molars and 21 molars, half maxillary and half
mandibular) were used in this study. Teeth were
placed three in a row (two premolars and one molar
or one premolar and two molars) with proximal
surfaces in contact and roots embedded in wax and
mounted in dental stone models. In total, 84 proximal
and 42 occlusal surfaces were available for evaluation.
Teeth were radiographed to make conventional ®lm
images and DR images. A Heliodent MD dental X-ray
unit (Siemens, Bensheim, Germany) operating at 7 mA
and 70 kVcp was used for all exposures. A 2 cm
thickness of tissue equivalent material (model 501A,
Radiation Measurements Inc., Middleton, WI, USA)
was used to simulate soft tissues. The projection
geometry was not standardized in order to simulate
real clinical conditions. However, a source-to-image
receptor distance of about 26 cm was maintained
throughout the experiment. Exposure times were
selected individually for each system (®lm or DR) to
provide adequate image densities.
Film images were made with No. 2-sized Ektaspeed
Plus ®lm (Eastman Kodak, Rochester, NY, USA) at
0.16 s. The projection geometry used to obtain ®lm
images simulated the bitewing technique, with two
models of three teeth each being exposed on the same
®lm. Films were processed in a Dent-X 810 automatic
processor (Dent-X, Elmsford, NY, USA) at 288C, with
a 4.5 min processing cycle. Radiographs were mounted
in opaque plastic holders and coded for later use.
DR images were obtained with a No. 2 CDR charge-
coupled device sensor (Schick Technologies Inc., Long
Island City, NY, USA) using an exposure time of 0.06 s.
For those images that would later be displayed as
unaltered DR images, only one projection was acquired
2- and 3-D imaging for caries
M Abreu et al
153
using a bitewing projection geometry. In the case of
images that would later undergo TACT reconstruction,
a 1 mm lead sphere (X-Spot, Beekley Corporation,
Bristol, CT, USA) was placed on the facial aspect of
each tooth at its cervical region and eight di€erent
source projections were obtained using di€erent vertical
and horizontal angulations. The di€erent positions of
the X-ray source were controlled manually, and there-
fore were not perfectly in the same plane. However, the
object-receptor relationship was kept constant, so that
the lead sphere could be used as a ®ducial marker when
reconstructing the TACT images.7
All digital images were exported to an IBM-
compatible personal computer (Paci®c Computers,
Chapel Hill, NC, USA). TACT Workbench software
(Verity Software Systems, Winston-Salem, NC, USA)
was used to generate the TACT images. Two kinds of
TACT images were generated for each tooth:
iteratively restored slices (TACT-S) and pseudoholo-
grams (TACT-PH). For TACT-S, a set of 12 ± 20
iteratively restored mesio-distal slices of each tooth was
prepared for viewing as a sequential stack. A total of
three iterative restorations were performed. TACT slice
generation and iterative restoration have been de-
scribed elsewhere.7,12 For TACT-PH, 72 two-dimen-
sional image frames were generated, representing
multiple views of the same tooth from di€erent
angles, and displayed in a sequential manner con-
trolled by the viewer. In a pilot study, iterative
restoration did not improve TACT-PH image quality
and was therefore not used. After TACT reconstruc-
tions, four di€erent image modalities of each tooth
were available for the observers to evaluate: conven-
tional ®lm, DR, and TACT-S (Figure 1), and TACT-
PH (Figure 2).
Viewing sessions
Eight observers were recruited for this study. They
were asked to score the presence or absence of caries in
the proximal and occlusal surfaces of the teeth along
with their con®dence in the assessment. This was
achieved using a 5-point con®dence rating scale in
which 1=caries de®nitely absent, 2=caries probably
absent, 3=unsure if caries absent or present, 4=caries
probably present, and 5=caries de®nitely present. To
avoid bias, observers were informed that the likelihood
of caries occurrence was 50% for any surface under
examination, and that they should consider caries as
any decalci®cation of the tooth surface.
Observers viewed images in four di€erent viewing
sessions, one for each image modality. The sequence of
modalities presented to the observers was system-
atically arranged in a way that, on average, no image
modality was seen earlier than any other. At least 1
week was allowed to elapse between consecutive
viewing sessions. All viewing sessions took place in a
quiet room with subdued ambient lighting. To ensure
reader calibration, a training exercise was completed
before each observation session, where instructions
were provided and observers became familiar with the
image modalities to be evaluated.
 2- and 3-D imaging for caries
M Abreu et al
154

Figure 1 Representative images of ®lm, DR and TACT-S used in this study. Caries status (maximum depth) of this tooth, determined by
histology, was as follows: left proximal surface ± caries extending to the inner third of dentin; occlusal surface ± caries extending to the inner half
of enamel; right proximal surface ± caries extending to the middle third of dentin

Table 1 Distribution of occlusal and proximal surfaces of 20

premolar and 20 molar teeth by caries status assessed by histologic
examination
Occlusal Proximal
n (%) n (%)
No decalcification 17 (42.5) 37 (46.25)
Decalcification in the external 0 (0) 19 (23.75)
half of enamel
Decalcification to the internal 7 (17.5) 7 (8.75)
half of enamel
Decalcification to the external 4 (10) 11 (13.75)
third of dentin
Decalcification to the middle 5 (12.5) 5 (6.25)
third of dentin
Decalcification to the internal 7 (17.5) 1 (1.25)
third of dentin
Total 40 (100) 80 (100)
Figure 2 Four of 72 TACT-PH image frames of the same tooth seen
in Figure 1. Notice how the perceived viewing angle changes on each
frame and how the visibility of the proximal surface lesions change
with di€erent angles

randomly selected without replacement so that the

Film images were viewed on a conventional viewbox.
Masks and a 26-magnifying viewer were available for
observers to use. Digital images (DR, TACT-S and
TACT-PH) were viewed on a 17-inch high-resolution
(SVGA) color monitor (Trinitron Multiscan 17sf II,
Sony Electronics, USA) using TACT Workbench
Software as the interface. The monitor resolution was
set at 8006600 pixels with a 16-bit color depth. Prior
to the viewing sessions, two board-certi®ed oral and
maxillofacial radiologists subjectively determined an
optimal contrast and brightness setting for each image.
To limit the number of variables, these settings were
kept constant throughout the experiment. Observers
were not allowed to enhance or modify the image
display in any other way. When either TACT modality
was being viewed, the observer was instructed to use
the `page-up/page-down' keys of the keyboard to
evaluate all available slices in the stack or frames in
the pseudohologram respectively.
To assess intra-observer reliability, a ®fth viewing
session was scheduled at least 1 week after the fourth
session. On that occasion, observers viewed 15% of the
images from each modality a second time. The
sequence in which modalities were presented to
observers was once again balanced. Images were
observer viewed a particular surface with just one
modality during the session.
Ground truth and statistical analysis
Subsequent to imaging, teeth were sectioned mesiodis-
tally in approximately 300 mm-thick sections using a
low speed saw equipped with a diamond blade (Buehler
Ltd., Lake Blu€, IL, USA). Two teeth, one premolar
and one molar, were accidentally destroyed during
sectioning. Sections were examined under a dissecting
microscope for the presence or absence of caries. If
present, the maximum depth was scored as the caries
status of that surface. Table 1 shows the distribution of
occlusal and proximal surfaces by caries status
determined by histological examination.
Observers' assessments with each image modality
were compared with the ground truth to determine
diagnostic performance in terms of areas under
Receiver Operating Characteristic (ROC) curves (Az).
Az values were analysed using an analysis of variance
(ANOVA) to test the main e€ects of modality,
observer, and surface, and the interactions between
(1) surface and modality and (2) observer and
modality. The level of signi®cance was set at a=0.05.
Intraclass correlation (ICC) was used to assess intra-
and inter-observer reliability.
 2- and 3-D imaging for caries
M Abreu et al
155
Results

Tables 2 and 3 show individual and mean Az values

calculated for the eight observers detecting occlusal
and proximal caries with the four image modalities. In
the detection of occlusal caries, mean Az values for
®lm, DR, TACT-S and TACT-PH were 0.77, 0.72,
0.75, and 0.68 respectively and for proximal caries were
0.78, 0.77, 0.76, and 0.74 respectively. Variability in
observer performance can be noticed from the range of
Az values found in the same modality, most marked for
occlusal caries. Figures 3 and 4 show ROC curves
representing the mean Az values of the eight observers
using the four image modalities for the detection of
occlusal and proximal caries, respectively. There were
no statistically signi®cant di€erences among observers
(P=0.276), modalities (P=0.310) or tooth surfaces
(P=0.195), as well as no signi®cant interactions
between (1) surface and modality (P=0.852) and (2)
observer and modality (P=0.838) (Table 4). o
ICC assessments of intra-observer variability
demonstrated that observers were, on average, more
Figure 3 ROC curves representing the mean Az values of eight
consistent in the scores given for ®lm than any of observers using the four image modalities for the detection of
the other three modalities (Table 5). Agreement was, occlusal caries
on average, higher for ®lm, DR and TACT-S than
for TACT-PH. Similarly, intra-observer ICC coeffi-
cients demonstrated that there was more agreement
among observers for ®lm than the other modalities
(Table 6).

Table 2 Areas under the ROC curve (Az) and standard deviations
(s.d.) by image modality for the eight observers in the detection of
occlusal caries
Occlusal caries detection
Film Digital TACT-S TACT-PH
Observer Az (s.d.) Az (s.d.) Az (s.d.) Az (s.d.)
1 0.83 (0.07) 0.85 (0.07) 0.72 (0.08) 0.78 (0.08)
2 0.79 (0.10) 0.82 (0.08) 0.71 (0.09) 0.65 (0.16)
3 0.71 (0.18) 0.58 (0.16) 0.87 (0.16) 0.58 (0.22)
4 0.77 (0.09) 0.64 (0.09) 0.77 (0.08) 0.72 (0.09)
5 0.89 (0.08) 0.61 (0.10) 0.58 (0.10) 0.70 (0.14)
6 0.84 (0.11) 0.79 (0.10) 0.78 (0.16) 0.79 (0.14)
7 0.59 (0.14) 0.55 (0.14) 0.84 (0.18) 0.37 (0.11)
8 0.74 (0.09) 0.89 (0.05) 0.72 (0.09) 0.83 (0.07)
Mean 0.77 (0.11) 0.72 (0.10) 0.75 (0.12) 0.68 (0.13)

Figure 4 ROC curves representing the mean Az values of eight

Table 3 Areas under ROC curves (Az) and standard deviations observers using the four image modalities for the detection of
(s.d.) by image modality for eight observers in the detection of proximal caries
proximal caries
Proximal caries detection
Film Digital TACT-S TACT-PH
Observer Az (s.d.) Az (s.d.) Az (s.d.) Az (s.d.)
Table 4 Analysis of variance (ANOVA)
1 0.72 (0.07) 0.64 (0.07) 0.76 (0.06) 0.74 (0.06)
2 0.73 (0.09) 0.84 (0.06) 0.78 (0.07) 0.81 (0.08) Sum of Mean
3 0.68 (0.11) 0.79 (0.07) 0.80 (0.09) 0.73 (0.10) Source squares df square f-ratio P
4 0.77 (0.08) 0.67 (0.09) 0.70 (0.08) 0.73 (0.08) Observer 0.094 7 0.013 1.323 0.276
5 0.81 (0.06) 0.70 (0.08) 0.81 (0.05) 0.68 (0.09) Modality 0.038 3 0.013 1.252 0.310
6 0.84 (0.06) 0.80 (0.07) 0.75 (0.07) 0.69 (0.07) Surface 0.018 1 0.018 1.759 0.195
7 0.84 (0.11) 0.87 (0.06) 0.67 (0.11) 0.75 (0.07) Surface*Modality 0.008 3 0.003 0.263 0.852
8 0.84 (0.05) 0.84 (0.06) 0.83 (0.06) 0.78 (0.07) Observer*Modality 0.139 21 0.007 0.657 0.838
Mean 0.78 (0.08) 0.77 (0.07) 0.76 (0.07) 0.74 (0.08) Error 0.283 28 0.010
 2- and 3-D imaging for caries
M Abreu et al
156
Table 5 Individual and mean intra-observer intraclass correlation receptor was used, and the source projections required
(ICC) coefficients by modality for the TACT reconstructions were obtained using
Intra-observer ICC tightly controlled projection geometry and a specially
Observer Film Digital TACT-S TACT-PH designed eight-tube X-ray source. Although we used a
1 0.72 0.46 0.57 0.79 current-generation CCD sensor, our protocol called for
2 0.64 0.24 0.40 0.61 hand-controlled angulations using a common X-ray
3 0.84 0.80 0.64 0.51 unit to simulate more closely a real clinical application
4 0.82 0.82 0.73 0.80
5 0.48 0.65 0.46 0.17 of TACT. The use of the same sensor together with
6 0.78 0.66 0.80 0.03 more stringent projection geometry may have produced
7 0.86 0.90 0.90 0.63 di€erent results. This speculation is justi®able if we
8 0.80 0.55 0.85 0.02 consider that the theory underlying TACT recommends
Mean* 0.74 0.67 0.67 0.46
that source projections be obtained from source-point
*Mean ICC coecients calculated from the mean-square values positions (position of the X-ray focal spot at each
derived from analysis of variance of the two scores given by the
observer to the same surface projection) that are in a same plane in relation to the
image receptor or are located far enough away from it
to avoid signi®cant image magni®cation.7 Accordingly,
those studies which reported favorable results for
Table 6 Interobserver intraclass correlation (ICC) coefficients by TACT in comparison to other modalities used
modality machine-controlled projection geometries.8 ± 11 The
Interobserver ICC source of the images used for TACT reconstructions
Film Digital TACT-S TACT-PH
in this study may not be ideal for the task of caries
Mean square surface 12.68 11.57 10.95 10.90 detection.
Mean square error 0.96 1.12 1.23 1.25 TACT is a relatively new technology. As such, its
ICC 0.60 0.54 0.50 0.49
capabilities are still being explored. Proprietary TACT
*ICC values derived from analysis of variance of the scores given by
the eight observers to the same surface
software o€ers a variety of algorithms and variables to
work with, providing many possible combinations of
settings.7 Similarly, the TACT radiographic data
acquisition scheme is subject to numerous variations
and few restrictions. The selection of the most
Discussion appropriate projection geometry and angular disparity
to be used is dependent on the diagnostic task. The
This in vitro study compared the ability of observers to number of source projections and iterative restorations
identify primary caries using four di€erent imaging used can also be varied to address the speci®c task at
modalities. Two of these modalities, conventional ®lm hand. So far, no unique combination of factors has
and DR, provide information in only two dimensions. been determined as the ideal for the di€erent diagnostic
The other two modalities, TACT-S and TACT-PH, are tasks encountered in dentistry. With a clinical
derived from a relatively new data-acquisition scheme application of TACT technology in mind, we used
that, with the aid of specialized software, renders a only one combination of factors to produce TACT
form of three-dimensional information to the viewer.7 images. Investigations are currently underway to
Theoretically, the three-dimensional capabilities might explore the various elements associated with image
be expected to impart superior diagnostic performance quality in TACT, trying to determine the optimal set of
over two-dimensional systems. This study, however, factors that should be used when performing common
was not able to demonstrate any signi®cant di€erences dental imaging tasks.
among the modalities tested. Fixed settings of brightness and contrast for digital
For all observers participating in this study, this was format images were selected for three reasons: ®rst,
their ®rst exposure to TACT images. Similarly, half of improperly used `enhancements' may actually decrease
the observers had very little experience with direct diagnostic performance,13 second, ®xed settings gave an
digital images. Even with training sessions, the equal change of comparable performance to those
observers' lack of familiarity with `computerized' observers less experienced with digital format images,
modalities may have played an important role in the and third, viewing time was shortened, reducing the
outcome of this study. The ICC coecents calculated risk of observer fatigue. The authors do not believe
for intra- and inter-observer variability substantiate that constraining contrast and brightness reduced the
this assumption, showing consistently more concor- performance of the 3D (or pseudo-3D) features of
dance for ®lm than the other three modalities. TACT images in this study.
In a previous study, Tyndall et al. found no This study is in agreement with other studies in the
signi®cant di€erence between the diagnostic perfor- literature showing that DR images are comparable
mances of ®lm and TACT images.12 Although our with ®lm in the detection of caries.4,13 Also, our results
results are in agreement, di€erent protocols were used. con®rm the large variation that has been reported in
In the earlier study, only the most central TACT slice observers' performance in diagnosing caries.14 This
was presented to the observers, instead of a complete variation among observers was in part responsible for
stack of slices as used in this study and pseudoholo- our inability to detect di€erences in the tested
grams were not evaluated. An obsolete digital image modalities.

In conclusion, this study was not able to demon-
strate signi®cant di€erences between ®lm, digital
radiography, iteratively restored TACT slices and
TACT pseudoholograms in the detection of caries.
The use of free-hand positioning of the X-ray source
References
1. Douglass CW, Valachovic RW, Wijesinha A, Chauncey HH,
Kapur KK, McNeil BJ. Clinical ecacy of dental radiography in
the detection of dental caries and periodontal disease. Oral Surg
Oral Med Oral Pathol 1986; 62: 330 ± 339.
2. Espelid I. Radiographic diagnosis and treatment decision on
approximal caries. Community Dent Oral Epidemiol 1986; 14:
265 ± 270.
3. Mùystad A, Svanaes DB, Risnes S, Larheim TA, GroÈndahl H-G.
Detection of approximal caries with a storage phosphor system.
A comparison of enhanced digital images with dental x-ray ®lm.
Dentomaxillofac Radiol 1996; 25: 202 ± 206.
4. Wenzel A. Digital radiography and caries diagnosis. Dentomax-
illofac Radiol 1998; 27: 3 ± 11.
5. Grant DG. Tomosynthesis: a three-dimensional radiographic
imaging technique. IEEE Trans Biomed Eng 1972; 19: 20 ± 28.
6. Groenhius RA, Webber RL, Ruttimann UE. Computerized
tomosynthesis of dental tissues. Oral Surg Oral Med Oral Pathol
1983; 56: 206 ± 214.
7. Webber RL, Horton RA, Tyndall DA, Ludlow JB. Tuned-
aperture computed tomography (TACTTM). Theory and
application for three-dimensional dento-alveolar imaging.
Dentomaxillofac Radiol 1997; 26: 53 ± 62.
8. Webber RL, Horton RA, Underhill TE, Ludlow JB, Tyndall
DA. Comparison of ®lm, direct digital, and tuned-aperture
computed-tomography (TACT) images for identifying the
location of crestal e€ects around endosseous titanium im-
plants. Oral Surg Oral Med Oral Pathol Oral Radiol Endod
1996; 81: 480 ± 490.
2- and 3-D imaging for caries
M Abreu et al
157
for acquisition of images for TACT reconstruction may
not be optimal for primary caries assessment. Further
research is needed to investigate the e€ects of imaging
variables on the diagnostic ecacy of TACT.
9. Horton RA, Ludlow JB, Webber RL, Gates W, Nason RH,
Reboussin D. Detection of peri-implant bone changes with axial
tomosynthesis. Oral Surg Oral Med Oral Pathol Oral Radiol
Endod 1996; 81: 124 ± 129.
10. Vandre RH, Webber RL, Horton RA, Cruz CA, Pajak JC.
Comparison of TACT with ®lm for detecting osseous jaw
lesions. J Dent Res 1994; 74 (special issue): 18.
11. Nair MK, Tyndall DA, Ludlow JB, May K. Tuned aperture
computed tomography and detection of recurrent caries. Caries
Res 1998; 32: 23 ± 30.
12. Tyndall DA, Clifton TL, Webber RL, Ludlow JB, Horton RA.
TACT imaging of primary caries. Oral Surg Oral Med Oral
Pathol Oral Radiol Endod 1997; 84: 214 ± 225.
13. Tyndall DA, Ludlow JB, Platin E, Nair M. A comparison of
Kodak Ektaspeed Plus ®lm and the Siemens Sidexis digital
imaging system for caries detection using receiver operating
characteristic analysis. Oral Surg Oral Med Oral Pathol Oral
Radiol Endod 1998; 85: 113 ± 118.
14. GroÈndahl H-G. The in¯uence of observer performance in
radiographic caries diagnosis. Swed Dent J 1979; 3: 101 ± 107.