- Email: [email protected]
Detection of mineral loss in approximal enamel by subtraction radiography
Vol. 77 No. 2
ORAL AND MAXILLOFACIAL
RADIOLOGY
February
1994
Editor: Allan G. Farman
Detection of mineral loss in approximal enamel by subtraction radiography Agnar Halse, Dr odont,a Ivar Espelid, Dr odont,b Anne Bj$rg Tveit, Dr odont,b and Stuart C. White, DDS, PhD,C Bergen, Norway and Los Angeles, Calif. SCHOOL OF DENTISTRY,
BERGEN AND UCLA SCHOOL OF DENTISTRY,
LOS ANGELES
The purpose of the present study was to determine whether digital subtraction radiography will improve detectability of small, mechanically prepared defects within dental enamel. Lesions with an extent of 1 mm in vertical direction and representing 5% to 10% mineral loss in the direction of the x-ray beam were prepared in eight extracted molars. Radiographs of teeth with defects were subtracted from radiographs taken before the lesions were prepared. Seven observers evaluated the images using a five-point confidence rating scale (receiver operating characteristic technique). Examination of the original radiographs showed increasing accuracy of radiographic interpretation with increasing mineral loss as judged from the areas beneath the receiver operating characteristic curves. The same observation was made using subtraction images with and without contrast enhancement. There was no indication that subtracted images provided better diagnostic validity than the original radiographs. In conclusion, subtraction images do not seem to improve the diagnosis of welldefined lesions within dental enamel. (ORAL SURC ORAL MED ORAL PATHOL 1994;77:177-82)
The accuracy of radiologic diagnosis of dental caries has been examined by several investigators.1-6In recent years many studies have beenperformed with the aim of improving radiographic interpretation through computerized technologies; these were reviewed by Wenzel.7 Among the challenges in radiologic diagnosis is the detection of small areasof destruction within cancellous bone; this task is made difficult by the complexity of the normal structure depicted over the alteration. It has been clearly shown that diagnostic accuracy can be considerably improved when these overlying structures can be removed by subtracting two images made at intervals but with the same geometry.*j 9 Dental enamel is relatively homogenous as observed on a radiograph. Some inhomogeneities are present, however, especially the gradient of decreasing optical density from the proximal surface to the Supported by grants from Sonja and Harry H. Benkows fund and A.S. Norsk Dental Depots fund. %chool of Dentistry, Bergen, and Visiting Professor UCLA School of Dentistry. bSchool of Dentistry, Bergen. ‘UCLA School of Dentistry. Copyright @ 1994 by Mosby-Year Book, Inc. 0030-4220/94/%3.00 + 0 7/16/52099
dentinoenamel junction. As shown by Pitts and Renson,lo there are also numerous localized areas characterized by density that is different from the surrounding tissue. This indicates that the subtraction of radiographs taken before and after the occurrence of a natural lesion or the creation of an artificial defect might improve the diagnostic accuracy. In an earlier study we found that subtraction of radiographs taken before and after the treatment of caries lesions with a stannous fluoride solution, which resulted in a small density increase,revealedalterations that could not be detected by direct comparison of the radiographs.” This observation also appeared very promising in the detection of other alterations by subtraction radiography. In particular, we are interested in the use of this method for the detection of incipient carious lesions. It is not clear, however, if this technique can be used effectively for such small defects in a relatively radiopaque object. Accordingly, the purpose of the present study was to determine whether subtraction will improve the detectability of small mechanically prepared defects within dental enamel. MATERIAL AND METHODS
The experimental material consisted of eight extracted permanent molars; all were clinically sound. 177
178
Hake et al.
Fig. 1. Molar cut longitudinally into threeparts.Approximal defectis preparedin the middle sectionthat extends from the prominenceto the dentinoenameljunction. Each tooth was sectioned vertically twice to produce a 0.5 to 0.7 mm thick mesiodistal longitudinal section (Fig. 1). The teeth with the section in position were mounted in methyl methacrylate tubes with elastic impression material. The tubes were fixed in a device that made it possibleto repeat radiographic exposures with identical projections. I2 Exposures were made at 65 kVp, the focus-object distance was 32.0 cm, and the object-film distance was 1.5 cm. A 1.O cm wide acrylic plastic container filled with water was placed between focus and object to simulate soft tissues. Kodak DF-58 (double-pack) (Eastman Kodak Co., Rochester, N.Y.) film was used. All radiographs were processed by a standardized automatic procedure. The sections were removed from the tubes, and an air-rotor with a l-mm cylindrical diamond bur was used to prepare lesions in the enamel from the most prominent part of the mesial or distal surface to the dentinoenamel junction (Fig. 1). In one molar two lesions were prepared. The lesions were about 1 mm wide. Then the sections were replaced between the buccal and lingual parts of the teeth, and new radiographs taken. The sections were removed, and the whole section was reduced by grinding about 100 pm in thickness, replaced, and radiographed until the lesions were just visible in the radiographs. Exposure parameters and geometry were the same in each exposure. The teeth were then embedded in methyl methacrylate and sectioned transversely through the lesions. To calculate the mineral loss in the x-ray beam
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY February 1994
direction, the thickness of each section, which had been measured with a micrometer caliper, was compared with the width of its tooth. This dimension was measured at an arbitrary depth of 1 mm below the lesion border, approximately corresponding to the enamel thickness in this region, in a direction parallel to the central ray of the x-ray beam. The radiographs were categorized on the basis of the measurementsinto three groups according to the calculated mineral loss: 5% mineral loss (range, 3.6% to 5.9%); 7% mineral loss (range, 6.1% to 7.9%); 10% mineral loss (range, 8.O%to 13.4%). The radiographs were digitized and subtracted as described previously. *’ After subtracting the 5%, 7%, and 10% images from the pretreatment images, the subtracted images were photographed from the screen by Kodak Rapid Process Copying film (Eastman Kodak Co.). The subtracted images were also contrast enhanced by interactively manipulating the displayed contrast modified with a linear end stretch algorithm. Two of us (A.H. and S.C.W.) decided jointly on the optimal level of contrast to view the actual region. Another photo was then made. The projected slides were judged to be essentially identical with the comparable images viewed on the screen. The image material to be evaluated now consisted of 24 radiographs, 24 subtraction images, and 24 contrast-enhanced subtraction images. Nine of 16 approximal surfaces had lesions, whereas sevenwere intact. The images were judged first by four clinical instructors in oral radiology, all trained in caries diagnosis and familiar with the receiver operating characteristic (ROC) data collection system. The images were also evaluated by three of us (A.H., I.E., and A.B.T.). The instructors were informed about the purposeof the study. They were not previously trained in viewing subtraction images. A series of conventional and subtraction images were discussed,and the observers expressedconfidence in their ability to interpret this type of image. No information about the frequency of the prepared lesions was given. The observers who examined the photographs of the subtraction images were given the opportunity to select magnification and to vary the light intensity of the projector. They were provided a ~2 magnification viewer to view the original radiographs and allowed to select luminous intensity. Detection of the defects was recorded on a 5-point confidence scale: 0 = lesion definitely not present; 1 = lesion probably not present; 2 = equal chance of being present or not; 3 = lesion probably present; 4 = lesion definitely present. The scorings were classified into a so-called decision matrix and treated ac-
Hake et al.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY Volume 77, Number 2
779
cording to the ROC method.i3* l4 A maximum-likelihood estimate of a ROC-curve parameter was obtained with the use of a computer program developed by Centor and Keightley l5 that was based on the original program by Dorfman and Alf.16 Statistical testing was performed according to Hanley and McNeil.17 The results are presented as A, values, which represent the efficiency of the diagnostic method.
Table I. Lesion size and diagnostic validity of three imaging modalities expressedas mean (standard error) of A, values
RESULTS lesion size
For each ima,e type all A, values at 7% defect size are significantly larger (p < 0.05) than the corresponding 5% defect size values. For each image type all A, values at 10%defect size are significantly larger (p < 0.05) than the corresponding 7% defect size values.
The accuracy of detection of the defects increased with increasing defect size. This was true for all observers and all imaging modalities (Table I). Observers
No difference was demonstrated between the observers within each group of imaging modalities. (Table II). Three of us (observers 5,6, and 7) did not achieve higher validity than the clinical instructors in oral radiology (observers 1 to 4). Imaging modalities
For most observers, examination of the subtracted images with and without contrast enhancement demonstrated an accuracy at approximately the same level as the original radiographs. Observer 5 obtained a significantly lower (p < 0.05) accuracy when using the subtraction images. Within the experimental group with 5% mineral loss, interpretation of the radiographs was significantly better (p < 0.05) than that of the contrast-enhanced subtraction images. The other differences were not significant (Fig. 2). Interpretation of the different images of the lesions that represented 7% mineral loss showed no significant differences (Fig. 3). Within the group that represented 10% mineral loss there was a significant difference (p < 0.05) between the radiographs and the subtraction images, as well as between the radiographs and the contrast-enhanced subtraction images (Fig. 4). As shown in Tables I and II and in Figs. 2 to 4, all differences were small. DISCUSSION
The accuracy of the detection of the mechanically prepared enamel defects correlated well with increasing mineral loss. This is consistent with studies that show a higher degree of detectability on radiographs of large enamel carious lesions18*l9 and artificial lesionsZocompared with small ones. The working hypothesis of the study was that removal of structural noise by subtraction might produce a higher accuracy for the subtracted imagesthan for the radiographs. This hypothesis was not supported by the findings. In fact, when a difference was
Subtractions, Defect size 5%
7% 10%
Radiographs
Subtractions
contrast enhanced
0.70 (.03) 0.78 (.04) 0.90 (.02)
0.65 (.03) 0.75 (.02) 0.85 (.03)
0.66 (.04) 0.72 (.04) 0.85 (.04)
Table II. Diagnostic accuracy obtained by different observers with the use of the three imaging modalities expressedas the mean (standard error) of the A, values; all defect sizes are combined Observers 1 2 3 4 5* 6 7
Radiographs 0.81 0.80 0.78 0.77 0.83 0.76 0.79
(0.06) (0.06) (0.06) (0.06) (0.05) (0.06) (0.06)
Subtractions 0.77 0.74 0.70 0.72 0.70 0.79 0.78
(0.06) (0.06) (0.07) (0.06) (0.07) (0.06) (0.06)
Subtractions, contrast enhanced 0.70 0.77 0.75 0.69 0.65 0.77 0.77
(0.07) (0.06) (0.06) (0.07) (0.07) (0.06) (0.06)
*The A, value of observer 5 is significantly (p < 0.05) higher for the radiographs than for the subtractions.
found between the imaging modalities, it was in favor of the conventional radiographs. A few studies on the usability of subtraction for the detection of alteration in dental hard tissueshave been published. Halse et al.” as well as Wenzel and Halse” found that an uptake of stannous fluoride in natural carious lesions could be visualized by subtraction even in caseswhen the visual comparison of pre- and posttreatment images failed to disclose any difference. Maggio et a1.,22in a laboratory study that used subtraction to detect progression of dentinal caries, were able to reveal both an increase and a decreasein radiographic density at the border of the lesions. Nummikoski et a1.23examined defects adjacent to composite fillings and found that subtraction was superior to conventional radiography. In contrast to these investigators who reported an improvement in diagnosis, Hintze et a1.24in a study of small prepared root cavities found that accuracy was not increased compared with conventional radiography. A possible reason for these somewhat conflicting observations may be the differences in both the tissuesinvestigated and the procedures for creating the alterations. Several research groups have used digital subtrac-
180
Ha&e et al.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY February 1994
0.6
TRUE POSITIVE RATIO
0.6
-t-
RADIOGRAPHS
-o-
SUBTRACTION CONTRAST > ENHANCED
+ 0
0.1
0.2
0.3
SUBTRACTION 0.4
0.6
0.6
0.7
0.6
p=O.O28 NS NS
> 0.9
I
1
FALSE POSITIVE RATIO
Fig. 2. ROC curves for imaging methods in diagnosing enamel defects that represent 5% mineral loss. Data are from seven observers.
0.6
0.6
TRUE POSITIVE RATIO
0.5
RADIOGRAPHS
0.4
-@-
0.3
SUBTRACTION CONTRAST > ENHANCED \
0.2
NS
NS
0.1
* SUBTRACTION +-+.-L-+.-+--+---
0 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
+.-- +.. , 0.6
0.9
1
FALSE POSITIVE RATIO
Fig. 3. ROC curves for defects that represent 7% mineral loss. Data are from seven observers.
tion for diagnosisof the marginal periodontium.8gg,25-29 The investigations have clearly shown that the removal of structural noise produces a marked increase in the detectability of a small substance loss. The technique has also proved useful for detection of alteration in the apical periodontium.30q31In contrast to bone, lesions in dental enamel will have a relatively uniform background. There are, however, a few structural elements that can interfere with the detec-
tion of a lesion. A double contour of the outer surface is often present especially in molars. There is also a marked decrease in radiographic density from the tooth surface to the dentinoenamel junction. As discussedby Pitts and Rensonl” there are also variations in radiographic density when recording in an incisalcervical direction. On balance, however, it is clear that the structural complexity of bone is greater than enamel. Accordingly, the major benefit of the
Hake et al.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY Volume 77, Number 2
181
0.6
TRUE POSITIVE RATIO
0.6
-t-
RADIOGRAPHS
+
SUBTRACTION p=o.o35
+ 0
0.1
0.2
SUBTRACTION 0.3
0.4
0.6
0.6
0.7
0.6
0.9
1
FALSE POSITIVE RATIO
Fig. 4. ROC curves for defects that represent 10% mineral loss. Data are from seven observers.
subtraction technique for bone may not apply to enamel. Although subtraction radiography offers a potential improvement of diagnostic accuracy by removal of noise, there are technical artifacts that may offset this improvement. The main problem in a clinical setting is standardization of the projection geometry. Several studies32y33have investigated various methods for securing identical image geometry by means of intraoral filmholders. Jansen et a1.34advocated the use of bilateral bite blocks. Also extraoral devices have been tested and found promising.35 Although arrangements for obtaining identical geometry may be further refined, the removal and repositioning in a clinical or a simulated clinical situation will seldombe capable of securing congruence at the 0.1 mm level. In addition, the processof digitizing and displaying a radiograph will unavoidably reduce the resolution of the image from that in the original radiograph. In the present investigation the improvement obtained by removing structural noise was probably balanced or more than balanced by the addition of artifacts. The present results, that the conventional radiographs were superior to the subtraction procedure for detection of well-delinated lesions in enamel, could be considered contrary to previous findings. However, the lesions under investigation differed from one study
to another. The well-defined lesions created in the present study are characterized by sharp borders to the enamel, and this may differ from natural carious lesions’ 1,21*22and wedge-shapedlesions33used in the other studies.
In conclusion, this model allowed comparative evaluation of the diagnostic efficacy of three competing imaging systemsfor detection of defectsin enamel when both the size of the defect and the fraction of mineral loss were known. We found that detection of the defects improved as the fraction of mineral loss increases. Finally, in this system when there is comparatively low complexity of structural detail, subtraction radiographs showed no advantage over conventional radiographs. REFERENCES 1. Marthaler T, Germann M. Radiographic and visual appearance of small smooth surface caries lesions studied on extracted teeth. Caries Res 1970;4:224-42. 2. Purdell-Lewis D, Groeneveld A, Pot T, Kwant G. Proximal carious lesions: a comparison of visual, radiographical, and microradiographical appearance. Ned Tijdschr Tandheelkd 1974;80 Neth Dent J (suppl 10):6-l% 3. Bille J, Thylstrup A. Radiographic diagnosis and clinical tissue changes in relation to treatment of approximal carious lesions. Caries Res 1982;16:1-6. 4. Mileman P. Radiographic caries diagnosis and restorative treatment decision making [PhD dissertation]. University of Groningen, Groningen, 1985. 5. Mejare I, Griindahl H-G, Carlstedt K, Grever A-C, Ottoson E. Accuracy at radiography and probing for the diagnosis of proximal caries. Stand J Dent Res 1985;93:178-84. 6. Espelid I, Tveit A. Clinical and radiographic assessment of approximal carious lesions. Acta Odontol Stand 1986;44:3 l-7. I. Wenzel A. Influence of computerized information technologies on image quality in dental radiographs. Dan Dent J 1991; 951527-59. 8. Grondahl H-G, Grijndahl K. Subtraction radiography for the diagnosis of periodontal bone lesions. ORAL SURC ORAL MED ORAL PATHOL
1983;55:208-13.
9. Hausmann E, Christersson R, Dunford R, Wikesjo U, Phyo J,
182
Hake
et al.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLCKY Frhruar~
IO.
11.
12. 13. 14.
15. 16.
17.
18.
19.
20.
21
22
23
24
Genco RJ. Usefulness of subtraction radiography in the evaluation of periodontal therapy. J Periodontol 1985:56(suppl I I ):4-7. Pitts NB, Renson CE. Further development of a computeraided image analysis method of quantifying radiolucencies in approximal enamel. Caries Res 1986;20:361-70. Halse A, White SC, Espelid I, Tveit AB. Visualization of stannous fluoride treatment of carious lesions by subtraction radiography. ORAL SURG ORAL MEU ORAL. PATHOI. 1990: 69:378-g I. Forsberg J. A method for experimental dental radiography. Int Endodon J 1992;25:93-6. Swets JA. The relative operating characteristic in psychology. Science 1973;182:990-1000. Swets JA, Picket RM. Evaluation of diagnostic systems: methods from signal detection theory. New York: Academic Press, 1982: 1.5-45. Centor RM, Keightley F. The ROC analyzer (Version 5.0). Richmond, Va., 1988. Dorfman DD, Alf E. Maximum-likelihood estimation of parameters of signal detection theory and determination of confidence intervals-rating-method data. J Math Psycho1 1969;6:487-96. Hartley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29-36. Espelid 1. Radiographic diagnosis and treatment decisions on approximal caries. Community Dent Oral Epidemiol 1986; 14:265-70. Espelid I, Tveit AB. Diagnostic quality and observer variation in radiographic diagnosis of approximal caries. Acta Odontol Stand 1986;44:39-46. Arnold LV. The radiographic detection of initial carious lesions on the proximal surfaces of teeth, [Thesis]. University of Utrecht Dental School, University of Groningen, The Netherlands, 1983. Wenzel A, Halse A. Digital subtraction radiography after stannous fluoride treatment for occlusal caries diagnosis. ORAI. SURG ORAL MED ORALPATHOL 1992;74:824-8. Maggio JJ, Hausman EM, Allen K, Potts TV. A model for dentinal caries progression by digital subtraction radiography. .I Prosthet Dent 1990;64:727-32. Nummikoski PV, Martinez TS, Matteson SR, McDavid WD, Dove SB. Digital subtraction radiography in artificial recurrent caries detection. Dentomaxillofac Radio1 1992;21:59-64. Hintze H, Wenzel A, Andreasen FM, Sewerin I. Digital subtraction radiography for assessment of simulated root resorp-
25.
26.
27.
28.
29.
30.
31.
32.
33. 34.
35
1994
tion cavities: performance of conventional and reverse contrast modes. Endod Dent Traumatol 1992%: 149-54. Jansen PTM. van Palenstein Helderman WH,van Akcn J. fhe detection of in vitro produced periodontal bone lesions h: conventional radiography and photographic subtraction radiography using observers and quantitative digital subtraction radiography. J Clin Periodontol I989;16:335-41. Okano T, Ohki M, Mera T, Soejima H, Ishikawa I. Yamada N. Quantitative evaluation of proximal bone lesions using digital subtraction radiography. Dentomaxillofacial Radio1 1988;17:99-103. Putnins E. Lavelle CLB, Holthuis A. Detection of three-walled infrabony defects by subtraction radiography. 0~41. SLKG 0RAl Mm ORAL PATHOL 1988;65:102-8. Schmidt EF. Webber RL. Ruttimann UE, Loesche WJ. Effect of periodontal therapy on alveolar bone as measured by subtraction radiography. J Periodont Res 1988;59:633-8. WenLel A, Warrer K, Karring T. Digital subtraction radiography in assessing bone changes in periodontal defects following guided tissue regeneration. .I Clin Periodontol 1992; I9:20813. Pascon EA, Introcaso JH, Langeland K. Development of predictable periapical lesion monitored by subtraction radiography. Endod Dent Traumatol 1987;3:192-208. Kullendorff 8, Griindahl K, Rohlin M, Henrikson CO. Subtraction radiography for the diagnosis of periapical bone lesions. Endod Dent Traumatol 1988;4:253-9. Rudolph DJ, White SC, Mankovich NJ. Influence of geometric distortion and exposure parameters on sensitivity of digital subtraction radiography. 0~41 &KG ORAL MED ORAL P,4THOl 1987;64:631-7. Sewerin I. Device for serial intraoral radiography with controlled projection angles. Dan Dent J 1990;94:613-7. Janssen PTM, van Palenstein Helderman WH. van Aken J. The effect of in-vivo-occurring errors in the reproducibility of radiographs on the use of the subtraction technique. J Clin Periodontol 1989;16:53-8. Jeffcoat MK. Reddy MS, Webber RL, Williams RC, Ruttimann UE. Extraoral control of geometry for digital subtraction radiography. J Periodont Res 1987:22:396-402.
Reprint requests. Agnar Halse, Dr odont School of Dentistry Arstadveien 17 N-5009 Bergen, Norway
ORAL AND MAXILLOFACIAL
RADIOLOGY
February
1994
Editor: Allan G. Farman
Detection of mineral loss in approximal enamel by subtraction radiography Agnar Halse, Dr odont,a Ivar Espelid, Dr odont,b Anne Bj$rg Tveit, Dr odont,b and Stuart C. White, DDS, PhD,C Bergen, Norway and Los Angeles, Calif. SCHOOL OF DENTISTRY,
BERGEN AND UCLA SCHOOL OF DENTISTRY,
LOS ANGELES
The purpose of the present study was to determine whether digital subtraction radiography will improve detectability of small, mechanically prepared defects within dental enamel. Lesions with an extent of 1 mm in vertical direction and representing 5% to 10% mineral loss in the direction of the x-ray beam were prepared in eight extracted molars. Radiographs of teeth with defects were subtracted from radiographs taken before the lesions were prepared. Seven observers evaluated the images using a five-point confidence rating scale (receiver operating characteristic technique). Examination of the original radiographs showed increasing accuracy of radiographic interpretation with increasing mineral loss as judged from the areas beneath the receiver operating characteristic curves. The same observation was made using subtraction images with and without contrast enhancement. There was no indication that subtracted images provided better diagnostic validity than the original radiographs. In conclusion, subtraction images do not seem to improve the diagnosis of welldefined lesions within dental enamel. (ORAL SURC ORAL MED ORAL PATHOL 1994;77:177-82)
The accuracy of radiologic diagnosis of dental caries has been examined by several investigators.1-6In recent years many studies have beenperformed with the aim of improving radiographic interpretation through computerized technologies; these were reviewed by Wenzel.7 Among the challenges in radiologic diagnosis is the detection of small areasof destruction within cancellous bone; this task is made difficult by the complexity of the normal structure depicted over the alteration. It has been clearly shown that diagnostic accuracy can be considerably improved when these overlying structures can be removed by subtracting two images made at intervals but with the same geometry.*j 9 Dental enamel is relatively homogenous as observed on a radiograph. Some inhomogeneities are present, however, especially the gradient of decreasing optical density from the proximal surface to the Supported by grants from Sonja and Harry H. Benkows fund and A.S. Norsk Dental Depots fund. %chool of Dentistry, Bergen, and Visiting Professor UCLA School of Dentistry. bSchool of Dentistry, Bergen. ‘UCLA School of Dentistry. Copyright @ 1994 by Mosby-Year Book, Inc. 0030-4220/94/%3.00 + 0 7/16/52099
dentinoenamel junction. As shown by Pitts and Renson,lo there are also numerous localized areas characterized by density that is different from the surrounding tissue. This indicates that the subtraction of radiographs taken before and after the occurrence of a natural lesion or the creation of an artificial defect might improve the diagnostic accuracy. In an earlier study we found that subtraction of radiographs taken before and after the treatment of caries lesions with a stannous fluoride solution, which resulted in a small density increase,revealedalterations that could not be detected by direct comparison of the radiographs.” This observation also appeared very promising in the detection of other alterations by subtraction radiography. In particular, we are interested in the use of this method for the detection of incipient carious lesions. It is not clear, however, if this technique can be used effectively for such small defects in a relatively radiopaque object. Accordingly, the purpose of the present study was to determine whether subtraction will improve the detectability of small mechanically prepared defects within dental enamel. MATERIAL AND METHODS
The experimental material consisted of eight extracted permanent molars; all were clinically sound. 177
178
Hake et al.
Fig. 1. Molar cut longitudinally into threeparts.Approximal defectis preparedin the middle sectionthat extends from the prominenceto the dentinoenameljunction. Each tooth was sectioned vertically twice to produce a 0.5 to 0.7 mm thick mesiodistal longitudinal section (Fig. 1). The teeth with the section in position were mounted in methyl methacrylate tubes with elastic impression material. The tubes were fixed in a device that made it possibleto repeat radiographic exposures with identical projections. I2 Exposures were made at 65 kVp, the focus-object distance was 32.0 cm, and the object-film distance was 1.5 cm. A 1.O cm wide acrylic plastic container filled with water was placed between focus and object to simulate soft tissues. Kodak DF-58 (double-pack) (Eastman Kodak Co., Rochester, N.Y.) film was used. All radiographs were processed by a standardized automatic procedure. The sections were removed from the tubes, and an air-rotor with a l-mm cylindrical diamond bur was used to prepare lesions in the enamel from the most prominent part of the mesial or distal surface to the dentinoenamel junction (Fig. 1). In one molar two lesions were prepared. The lesions were about 1 mm wide. Then the sections were replaced between the buccal and lingual parts of the teeth, and new radiographs taken. The sections were removed, and the whole section was reduced by grinding about 100 pm in thickness, replaced, and radiographed until the lesions were just visible in the radiographs. Exposure parameters and geometry were the same in each exposure. The teeth were then embedded in methyl methacrylate and sectioned transversely through the lesions. To calculate the mineral loss in the x-ray beam
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY February 1994
direction, the thickness of each section, which had been measured with a micrometer caliper, was compared with the width of its tooth. This dimension was measured at an arbitrary depth of 1 mm below the lesion border, approximately corresponding to the enamel thickness in this region, in a direction parallel to the central ray of the x-ray beam. The radiographs were categorized on the basis of the measurementsinto three groups according to the calculated mineral loss: 5% mineral loss (range, 3.6% to 5.9%); 7% mineral loss (range, 6.1% to 7.9%); 10% mineral loss (range, 8.O%to 13.4%). The radiographs were digitized and subtracted as described previously. *’ After subtracting the 5%, 7%, and 10% images from the pretreatment images, the subtracted images were photographed from the screen by Kodak Rapid Process Copying film (Eastman Kodak Co.). The subtracted images were also contrast enhanced by interactively manipulating the displayed contrast modified with a linear end stretch algorithm. Two of us (A.H. and S.C.W.) decided jointly on the optimal level of contrast to view the actual region. Another photo was then made. The projected slides were judged to be essentially identical with the comparable images viewed on the screen. The image material to be evaluated now consisted of 24 radiographs, 24 subtraction images, and 24 contrast-enhanced subtraction images. Nine of 16 approximal surfaces had lesions, whereas sevenwere intact. The images were judged first by four clinical instructors in oral radiology, all trained in caries diagnosis and familiar with the receiver operating characteristic (ROC) data collection system. The images were also evaluated by three of us (A.H., I.E., and A.B.T.). The instructors were informed about the purposeof the study. They were not previously trained in viewing subtraction images. A series of conventional and subtraction images were discussed,and the observers expressedconfidence in their ability to interpret this type of image. No information about the frequency of the prepared lesions was given. The observers who examined the photographs of the subtraction images were given the opportunity to select magnification and to vary the light intensity of the projector. They were provided a ~2 magnification viewer to view the original radiographs and allowed to select luminous intensity. Detection of the defects was recorded on a 5-point confidence scale: 0 = lesion definitely not present; 1 = lesion probably not present; 2 = equal chance of being present or not; 3 = lesion probably present; 4 = lesion definitely present. The scorings were classified into a so-called decision matrix and treated ac-
Hake et al.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY Volume 77, Number 2
779
cording to the ROC method.i3* l4 A maximum-likelihood estimate of a ROC-curve parameter was obtained with the use of a computer program developed by Centor and Keightley l5 that was based on the original program by Dorfman and Alf.16 Statistical testing was performed according to Hanley and McNeil.17 The results are presented as A, values, which represent the efficiency of the diagnostic method.
Table I. Lesion size and diagnostic validity of three imaging modalities expressedas mean (standard error) of A, values
RESULTS lesion size
For each ima,e type all A, values at 7% defect size are significantly larger (p < 0.05) than the corresponding 5% defect size values. For each image type all A, values at 10%defect size are significantly larger (p < 0.05) than the corresponding 7% defect size values.
The accuracy of detection of the defects increased with increasing defect size. This was true for all observers and all imaging modalities (Table I). Observers
No difference was demonstrated between the observers within each group of imaging modalities. (Table II). Three of us (observers 5,6, and 7) did not achieve higher validity than the clinical instructors in oral radiology (observers 1 to 4). Imaging modalities
For most observers, examination of the subtracted images with and without contrast enhancement demonstrated an accuracy at approximately the same level as the original radiographs. Observer 5 obtained a significantly lower (p < 0.05) accuracy when using the subtraction images. Within the experimental group with 5% mineral loss, interpretation of the radiographs was significantly better (p < 0.05) than that of the contrast-enhanced subtraction images. The other differences were not significant (Fig. 2). Interpretation of the different images of the lesions that represented 7% mineral loss showed no significant differences (Fig. 3). Within the group that represented 10% mineral loss there was a significant difference (p < 0.05) between the radiographs and the subtraction images, as well as between the radiographs and the contrast-enhanced subtraction images (Fig. 4). As shown in Tables I and II and in Figs. 2 to 4, all differences were small. DISCUSSION
The accuracy of the detection of the mechanically prepared enamel defects correlated well with increasing mineral loss. This is consistent with studies that show a higher degree of detectability on radiographs of large enamel carious lesions18*l9 and artificial lesionsZocompared with small ones. The working hypothesis of the study was that removal of structural noise by subtraction might produce a higher accuracy for the subtracted imagesthan for the radiographs. This hypothesis was not supported by the findings. In fact, when a difference was
Subtractions, Defect size 5%
7% 10%
Radiographs
Subtractions
contrast enhanced
0.70 (.03) 0.78 (.04) 0.90 (.02)
0.65 (.03) 0.75 (.02) 0.85 (.03)
0.66 (.04) 0.72 (.04) 0.85 (.04)
Table II. Diagnostic accuracy obtained by different observers with the use of the three imaging modalities expressedas the mean (standard error) of the A, values; all defect sizes are combined Observers 1 2 3 4 5* 6 7
Radiographs 0.81 0.80 0.78 0.77 0.83 0.76 0.79
(0.06) (0.06) (0.06) (0.06) (0.05) (0.06) (0.06)
Subtractions 0.77 0.74 0.70 0.72 0.70 0.79 0.78
(0.06) (0.06) (0.07) (0.06) (0.07) (0.06) (0.06)
Subtractions, contrast enhanced 0.70 0.77 0.75 0.69 0.65 0.77 0.77
(0.07) (0.06) (0.06) (0.07) (0.07) (0.06) (0.06)
*The A, value of observer 5 is significantly (p < 0.05) higher for the radiographs than for the subtractions.
found between the imaging modalities, it was in favor of the conventional radiographs. A few studies on the usability of subtraction for the detection of alteration in dental hard tissueshave been published. Halse et al.” as well as Wenzel and Halse” found that an uptake of stannous fluoride in natural carious lesions could be visualized by subtraction even in caseswhen the visual comparison of pre- and posttreatment images failed to disclose any difference. Maggio et a1.,22in a laboratory study that used subtraction to detect progression of dentinal caries, were able to reveal both an increase and a decreasein radiographic density at the border of the lesions. Nummikoski et a1.23examined defects adjacent to composite fillings and found that subtraction was superior to conventional radiography. In contrast to these investigators who reported an improvement in diagnosis, Hintze et a1.24in a study of small prepared root cavities found that accuracy was not increased compared with conventional radiography. A possible reason for these somewhat conflicting observations may be the differences in both the tissuesinvestigated and the procedures for creating the alterations. Several research groups have used digital subtrac-
180
Ha&e et al.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY February 1994
0.6
TRUE POSITIVE RATIO
0.6
-t-
RADIOGRAPHS
-o-
SUBTRACTION CONTRAST > ENHANCED
+ 0
0.1
0.2
0.3
SUBTRACTION 0.4
0.6
0.6
0.7
0.6
p=O.O28 NS NS
> 0.9
I
1
FALSE POSITIVE RATIO
Fig. 2. ROC curves for imaging methods in diagnosing enamel defects that represent 5% mineral loss. Data are from seven observers.
0.6
0.6
TRUE POSITIVE RATIO
0.5
RADIOGRAPHS
0.4
-@-
0.3
SUBTRACTION CONTRAST > ENHANCED \
0.2
NS
NS
0.1
* SUBTRACTION +-+.-L-+.-+--+---
0 0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
+.-- +.. , 0.6
0.9
1
FALSE POSITIVE RATIO
Fig. 3. ROC curves for defects that represent 7% mineral loss. Data are from seven observers.
tion for diagnosisof the marginal periodontium.8gg,25-29 The investigations have clearly shown that the removal of structural noise produces a marked increase in the detectability of a small substance loss. The technique has also proved useful for detection of alteration in the apical periodontium.30q31In contrast to bone, lesions in dental enamel will have a relatively uniform background. There are, however, a few structural elements that can interfere with the detec-
tion of a lesion. A double contour of the outer surface is often present especially in molars. There is also a marked decrease in radiographic density from the tooth surface to the dentinoenamel junction. As discussedby Pitts and Rensonl” there are also variations in radiographic density when recording in an incisalcervical direction. On balance, however, it is clear that the structural complexity of bone is greater than enamel. Accordingly, the major benefit of the
Hake et al.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLOGY Volume 77, Number 2
181
0.6
TRUE POSITIVE RATIO
0.6
-t-
RADIOGRAPHS
+
SUBTRACTION p=o.o35
+ 0
0.1
0.2
SUBTRACTION 0.3
0.4
0.6
0.6
0.7
0.6
0.9
1
FALSE POSITIVE RATIO
Fig. 4. ROC curves for defects that represent 10% mineral loss. Data are from seven observers.
subtraction technique for bone may not apply to enamel. Although subtraction radiography offers a potential improvement of diagnostic accuracy by removal of noise, there are technical artifacts that may offset this improvement. The main problem in a clinical setting is standardization of the projection geometry. Several studies32y33have investigated various methods for securing identical image geometry by means of intraoral filmholders. Jansen et a1.34advocated the use of bilateral bite blocks. Also extraoral devices have been tested and found promising.35 Although arrangements for obtaining identical geometry may be further refined, the removal and repositioning in a clinical or a simulated clinical situation will seldombe capable of securing congruence at the 0.1 mm level. In addition, the processof digitizing and displaying a radiograph will unavoidably reduce the resolution of the image from that in the original radiograph. In the present investigation the improvement obtained by removing structural noise was probably balanced or more than balanced by the addition of artifacts. The present results, that the conventional radiographs were superior to the subtraction procedure for detection of well-delinated lesions in enamel, could be considered contrary to previous findings. However, the lesions under investigation differed from one study
to another. The well-defined lesions created in the present study are characterized by sharp borders to the enamel, and this may differ from natural carious lesions’ 1,21*22and wedge-shapedlesions33used in the other studies.
In conclusion, this model allowed comparative evaluation of the diagnostic efficacy of three competing imaging systemsfor detection of defectsin enamel when both the size of the defect and the fraction of mineral loss were known. We found that detection of the defects improved as the fraction of mineral loss increases. Finally, in this system when there is comparatively low complexity of structural detail, subtraction radiographs showed no advantage over conventional radiographs. REFERENCES 1. Marthaler T, Germann M. Radiographic and visual appearance of small smooth surface caries lesions studied on extracted teeth. Caries Res 1970;4:224-42. 2. Purdell-Lewis D, Groeneveld A, Pot T, Kwant G. Proximal carious lesions: a comparison of visual, radiographical, and microradiographical appearance. Ned Tijdschr Tandheelkd 1974;80 Neth Dent J (suppl 10):6-l% 3. Bille J, Thylstrup A. Radiographic diagnosis and clinical tissue changes in relation to treatment of approximal carious lesions. Caries Res 1982;16:1-6. 4. Mileman P. Radiographic caries diagnosis and restorative treatment decision making [PhD dissertation]. University of Groningen, Groningen, 1985. 5. Mejare I, Griindahl H-G, Carlstedt K, Grever A-C, Ottoson E. Accuracy at radiography and probing for the diagnosis of proximal caries. Stand J Dent Res 1985;93:178-84. 6. Espelid I, Tveit A. Clinical and radiographic assessment of approximal carious lesions. Acta Odontol Stand 1986;44:3 l-7. I. Wenzel A. Influence of computerized information technologies on image quality in dental radiographs. Dan Dent J 1991; 951527-59. 8. Grondahl H-G, Grijndahl K. Subtraction radiography for the diagnosis of periodontal bone lesions. ORAL SURC ORAL MED ORAL PATHOL
1983;55:208-13.
9. Hausmann E, Christersson R, Dunford R, Wikesjo U, Phyo J,
182
Hake
et al.
ORAL SURGERY ORAL MEDICINE ORAL PATHOLCKY Frhruar~
IO.
11.
12. 13. 14.
15. 16.
17.
18.
19.
20.
21
22
23
24
Genco RJ. Usefulness of subtraction radiography in the evaluation of periodontal therapy. J Periodontol 1985:56(suppl I I ):4-7. Pitts NB, Renson CE. Further development of a computeraided image analysis method of quantifying radiolucencies in approximal enamel. Caries Res 1986;20:361-70. Halse A, White SC, Espelid I, Tveit AB. Visualization of stannous fluoride treatment of carious lesions by subtraction radiography. ORAL SURG ORAL MEU ORAL. PATHOI. 1990: 69:378-g I. Forsberg J. A method for experimental dental radiography. Int Endodon J 1992;25:93-6. Swets JA. The relative operating characteristic in psychology. Science 1973;182:990-1000. Swets JA, Picket RM. Evaluation of diagnostic systems: methods from signal detection theory. New York: Academic Press, 1982: 1.5-45. Centor RM, Keightley F. The ROC analyzer (Version 5.0). Richmond, Va., 1988. Dorfman DD, Alf E. Maximum-likelihood estimation of parameters of signal detection theory and determination of confidence intervals-rating-method data. J Math Psycho1 1969;6:487-96. Hartley JA, McNeil BJ. The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 1982;143:29-36. Espelid 1. Radiographic diagnosis and treatment decisions on approximal caries. Community Dent Oral Epidemiol 1986; 14:265-70. Espelid I, Tveit AB. Diagnostic quality and observer variation in radiographic diagnosis of approximal caries. Acta Odontol Stand 1986;44:39-46. Arnold LV. The radiographic detection of initial carious lesions on the proximal surfaces of teeth, [Thesis]. University of Utrecht Dental School, University of Groningen, The Netherlands, 1983. Wenzel A, Halse A. Digital subtraction radiography after stannous fluoride treatment for occlusal caries diagnosis. ORAI. SURG ORAL MED ORALPATHOL 1992;74:824-8. Maggio JJ, Hausman EM, Allen K, Potts TV. A model for dentinal caries progression by digital subtraction radiography. .I Prosthet Dent 1990;64:727-32. Nummikoski PV, Martinez TS, Matteson SR, McDavid WD, Dove SB. Digital subtraction radiography in artificial recurrent caries detection. Dentomaxillofac Radio1 1992;21:59-64. Hintze H, Wenzel A, Andreasen FM, Sewerin I. Digital subtraction radiography for assessment of simulated root resorp-
25.
26.
27.
28.
29.
30.
31.
32.
33. 34.
35
1994
tion cavities: performance of conventional and reverse contrast modes. Endod Dent Traumatol 1992%: 149-54. Jansen PTM. van Palenstein Helderman WH,van Akcn J. fhe detection of in vitro produced periodontal bone lesions h: conventional radiography and photographic subtraction radiography using observers and quantitative digital subtraction radiography. J Clin Periodontol I989;16:335-41. Okano T, Ohki M, Mera T, Soejima H, Ishikawa I. Yamada N. Quantitative evaluation of proximal bone lesions using digital subtraction radiography. Dentomaxillofacial Radio1 1988;17:99-103. Putnins E. Lavelle CLB, Holthuis A. Detection of three-walled infrabony defects by subtraction radiography. 0~41. SLKG 0RAl Mm ORAL PATHOL 1988;65:102-8. Schmidt EF. Webber RL. Ruttimann UE, Loesche WJ. Effect of periodontal therapy on alveolar bone as measured by subtraction radiography. J Periodont Res 1988;59:633-8. WenLel A, Warrer K, Karring T. Digital subtraction radiography in assessing bone changes in periodontal defects following guided tissue regeneration. .I Clin Periodontol 1992; I9:20813. Pascon EA, Introcaso JH, Langeland K. Development of predictable periapical lesion monitored by subtraction radiography. Endod Dent Traumatol 1987;3:192-208. Kullendorff 8, Griindahl K, Rohlin M, Henrikson CO. Subtraction radiography for the diagnosis of periapical bone lesions. Endod Dent Traumatol 1988;4:253-9. Rudolph DJ, White SC, Mankovich NJ. Influence of geometric distortion and exposure parameters on sensitivity of digital subtraction radiography. 0~41 &KG ORAL MED ORAL P,4THOl 1987;64:631-7. Sewerin I. Device for serial intraoral radiography with controlled projection angles. Dan Dent J 1990;94:613-7. Janssen PTM, van Palenstein Helderman WH. van Aken J. The effect of in-vivo-occurring errors in the reproducibility of radiographs on the use of the subtraction technique. J Clin Periodontol 1989;16:53-8. Jeffcoat MK. Reddy MS, Webber RL, Williams RC, Ruttimann UE. Extraoral control of geometry for digital subtraction radiography. J Periodont Res 1987:22:396-402.
Reprint requests. Agnar Halse, Dr odont School of Dentistry Arstadveien 17 N-5009 Bergen, Norway