Objective: To determine a proper scanning resolution for digitizing bitewing radiographs in the detection of approximal caries. Methods: Fifty-two premolars and ...
ã
Dentomaxillofacial Radiology (2001) 30, 166 ± 171 2001 Nature Publishing Group. All rights reserved 0250 ± 832X/01 $15.00 www.nature.com/dmfr
Scanning resolution and the detection of approximal caries A Janhom*,1, FC van Ginkel1, JP van Amerongen2 and PF van der Stelt1 1
Department of Oral Radiology, Academic Centre for Dentistry Amsterdam (ACTA), Amsterdam, The Netherlands; 2Department of Cariology Endodontology Pedodontology, Academic Centre for Dentistry Amsterdam (ACTA), Amsterdam, The Netherlands
Objective: To determine a proper scanning resolution for digitizing bitewing radiographs in the detection of approximal caries. Methods: Fifty-two premolars and 48 molars were mounted in blocks and imaged on conventional ®lm (Ektaspeed Plus, Eastman-Kodak, Rochester, NY USA) simulating a bitewing projection. The 15 bitewing radiographs were then scanned with a ¯atbed scanner at three resolutions 150, 300 and 600 d.p.i. The digitized images were displayed in random order on a high-resolution cathode ray tube monitor. Ten observers assessed the caries status of 200 approximal surfaces. They scored lesion presence on a 5-point con®dence scale and depth on a 3-point scale. The observer's scores were compared with the results from a histological examination. Data were analysed using analysis of variance, by calculating signed observer error, absolute observer error and observer con®dence. Results: Lesion depth had a signi®cant eect on con®dence of lesion recognition. The main eect of resolution and the interaction between resolution and lesion depth were signi®cant. Pair-wise comparison showed a signi®cant dierence between resolutions in case of sound surfaces and surfaces with dentinal lesions for absolute error. The con®dence increased as the resolution increased but no signi®cant dierence was found between 300 and 600 d.p.i. The best score for depth estimation was obtained at the 300 d.p.i. scanning resolution. Conclusions: When bitewing radiographs are scanned with a ¯atbed scanner, a resolution of 300 d.p.i. seems the best choice. At this resolution the digital ®le size is manageable without signi®cant loss of the information necessary for caries diagnosis. Keywords: digital radiography, dental; radiographic image enhancement; dental caries; observer variation Introduction Digitizing existing radiographs will be an essential feature of dental practice in the transition period between the use of analog radiographs and the routine adoption of digital imaging. There are many ways to convert an analog ®lm into a digital image. The digital image can be obtained by using a CCD camera, a laser scanner or a ¯atbed scanner. A ¯atbed scanner is not expensive and can also be used for document scanning in the dental oce. Increasing the scanning resolution will increase visible details in the image. However, more disk space is needed for storage and a longer time for electronic transmission. It has been shown in medical radiography that lowcost ¯atbed scanner hardware and accompanying *Correspondence to: A Janhom, ACTA, Department of Oral Radiology, Louwesweg 1, 1066 EA, Amsterdam, The Netherlands Received 23 September 2000; accepted 12 February 2001
software could be used for teleradiology of CT scans of the head.1 A personal-computer-based scanner was shown to provide good quality images for teleradiology consultation.2 Few studies, however, have investigated the appropriate scanning resolution of radiographs for dierent diagnostic tasks in dentistry.3 ± 6 Kassebaum et al. found no signi®cant dierence in the diagnostic accuracy for caries of images with dierent pixel sizes.3 Ohki et al. found that images with a pixel size of 100 mm and 32 gray levels were acceptable for caries detection.6 However, they only asked the participants to assess the presence of lesions and did not take lesion depth into account. It remains questionable, therefore, if all lesion depths would be detected equally well at a particular resolution. A recent study of the compression of radiographs digitized with a ¯atbed scanner for image transmission showed that some combinations of compression rate and sampling resolution were better
Scanning resolution A Janhom et al
167
than others for the detection of speci®c anatomical features.7 We were unable to ®nd a study investigating the general eect of scanning resolution on caries detection. The aim of this study was, therefore, to determine the proper scanning resolution when using a ¯atbed scanner for bitewing radiographs in the diagnosis of the presence and depth of approximal caries. Material and methods Image acquisition Fifty-two extracted human premolars and 48 molars were used in this study. Based on radiographic examination and visual inspection the proximal surfaces of the teeth ranged from sound to a small cavity. We attempted to include a variety of caries depths and an equal distribution of carious and noncarious surfaces. Premolar teeth were grouped together and mounted in sets of four. Molars were mounted in groups of three. In total, 13 premolar blocks and sixteen molar blocks were completed. Two blocks at a time were placed in a jig to obtain a reproducible projection and radiographs were taken simulating bitewing projection using No. 2 size Ektaspeed Plus ®lm (Eastman Kodak, Rochester, NY, USA). One premolar block was used twice in order to have seven bitewing radiographs. A total of eight bitewing radiographs were made from molar blocks. A Heliodent MD (Siemens, Bensheim, Germany) dental X-ray machine was used at 60 kVp, 7 mA a FFD of 35 cm and rectangular collimation. A 12-mm thick soft tissue equivalent material (Mix-D)8,9 was placed between the X-ray source and the tooth blocks to simulate the scattering eect of soft tissues. The exposure time used for the premolar blocks was 0.32 s and 0.4 s for the molar blocks. Using these settings, the density of dentine close to the proximal enamel was on average 0.8 when measured with a densitometer (MacBeth TD 502, Newburg, NY, USA). This is considered clinically acceptable.10 The radiographs were developed in an automatic processor (DuÈrr-Dental XR 24; Durr-Dental, Bietingheim-Bissingen, Germany) at 278C, with a 5.4 min processing cycle. Examples of the bitewing radiographs is shown in Figure 1. The radiographs were masked and secured in ®lm mounts (Dentsply Rinn, Elgin, IL, USA). All radiographs were then scanned on a ¯atbed scanner (Agfa DuoScan T1200; Agfa, Mortsel, Belgium), using the manufacturer's software (Agfa Fotolook 3.0; Agfa, Mortsel, Belgium) at 8 bits resolution. Three dierent resolutions were subsequently used, 150, 300 and 600 d.p.i.. These resolutions were chosen because they are similar to those available in the settings of storage phosphor plate scanners for intra-oral dental radiography. The ®lms were placed in the upper center of the scanner. Tone curve gamma was set at 2.2. All digital image ®les were saved in the BMP ®le format.
Figure 1 Examples of simulated bitewing images at their dierent resolutions, 150, 300 and 600 d.p.i.
In total, 200 proximal surfaces (96 surfaces of molars and 104 surfaces of premolars) were included in the scanned images. Observation sessions Mirror images of seven premolar bitewings and eight molar bitewings were produced. A total of 90 scanned images (30 images at each resolution) were therefore available for viewing. Ten observers viewed one image at a time in random order in a dimly lit room. The observers were asked to view only the approximal surfaces on the right-hand side of each tooth and for this the left-hand images were viewed as the mirror image. The sequence for each resolution was also randomized. The images were viewed on a 17-inch SVGA monitor (Digital, Gumi, Korea; 10246768 pixels) at about 50 cm. The observers were either staff members (full-time or part-time) of the Department of Oral and Maxillofacial Radiology or the Department of Cariology, Endodontics and Pedodontics at The Academic Center for Dentistry Amsterdam. The Dentomaxillofacial Radiology
Scanning resolution A Janhom et al
168
observers were asked to score the presence of radiolucency on a 5-point con®dence scale ranging from 1=de®nitely no proximal radiolucency, 2=probably no proximal radiolucency, 3=uncertain, 4=probably proximal radiolucency, and 5=de®nitely proximal radiolucency. For each score of 4 or 5, they were asked to specify the depth of the lesion as follows: 1=radiolucency restricted to the ®rst half of enamel, 2=radiolucency in enamel and up to DEJ and 3=radiolucency into dentine. In order to obtain the caries status, the teeth were sectioned with a diamond bur in a mesio-distal direction close to the lesion and ground so that the widest extent of the lesion could be histologically examined. The histological examination was done under a stereomicroscope with 106 magni®cation (Wild 355110, Heerbrugg, Switzerland).11 A consensus of the histological results was achieved by two investigators (JPvA and AJ). A total of 29 surfaces were excluded from the study because some of the teeth were damaged or found to have root or occlusal caries. The ®nal material consisted of 171 approximal surfaces. The distribution of lesion depth based on the histological examination of the sectioned teeth for the premolars was: sound surfaces=29; caries in enamel (corresponding to radiographic depth score 1)=30; caries beyond DEJ but not further than 0.5 mm into dentine (corresponding to radiographic depth score 2)=21; and caries extended into dentine (corresponding to radiographic depth score 3)=11. The distribution of molar surfaces was: sound tooth surfaces=49; caries in enamel=10; caries beyond DEJ but not further than 0.5 mm into dentine=10; and caries into dentine=11. Data analysis The observations for each level of resolution and the histological depth were analysed with the aid of the Statistical Package for the Social Sciences (SPSS version 9.0 for Windows). The signed and absolute observer errors were calculated. The signed observer error is obtained by subtracting the actual lesion depth from the depth that the observer reported. For example, if the histological examination revealed an enamel (score 1) lesion but the observer reported a sound surface (score 0), the signed observer error in this case would be 71 (an under-estimation of the depth by the observer). A signed observer error score of 0 is the optimal score, obtained when the observer reports the same depth as the histologically measured depth. It is a measure of the amount of over- or underestimation of the depth of lesions. However, as a measure of observer competence it may be less adequate. If the over- and under-estimation are equal, then the ®nal average score would be 0. Therefore the absolute observer error which shows the magnitude of the error was used as well. The absolute observer error is the absolute value of the signed observer error. The observer con®dence score
Dentomaxillofacial Radiology
was also calculated. This score indicates the amount of con®dence expressed (0=uncertain, 1=probably right, 2=de®nitely right). A minus sign was added to the con®dence score when the observer was con®dent in the wrong direction. Multivariate analysis of variance (MANOVA) was carried out with the signed observer error, absolute observer error and observer con®dence as dependent variables. The levels of resolution (150, 300, and 600 d.p.i.), and the identity of the observer (1 ± 10) were entered as `within subjects' factors. In the analysis the lesion depth and element type (premolar, molar) were treated as `between subjects' factors. Results were considered signi®cant when P50.05. Results Table 1 summarizes the results of multivariate tests for all three dependent variables (signed and absolute observer error and observer con®dence) jointly. The lesion depth and resolution main eects (P=0.000) as well as the lesion depth by resolution interaction eect were (multivariately) signi®cant (P=0.001). As a univariate level the lesion depth main eect was signi®cant for all three dependent variables (P=0.000). Figure 2 illustrates the lesion depth main eect for all resolutions combined. The negative averages for the signed observer error whenever a lesion is present means that the depth of a lesion is generally underestimated. The positive average signed observer error for no lesion images indicates that for sound surfaces sometimes a lesion was reported. Average observer con®dence is positive for sound surfaces and dentinal lesion images. On average,
Table 1 Multivariate tests for all three dependent variables (signed observer error, absolute observer error, and observer confidence). The following variable names are used in the table: Type=type of tooth (premolar or molar), Truedept=lesion depth (no lesion, enamel, up to DEJ, or dentinal lesion), Resolution=scanning resolution (150, 300, or 600 d.p.i.) and Observer=observer 1 to 10 Fa
Effect
Sig.c b
Between Intercept 948.829 Subjects Type 1.048b Truedept 95.433 Type * Truedept 1.051 Within Resolution 6.327b Subjects Resolution * Type 2.577b Resolution * Truedept 2.432 Resolution * Type * Truedept 1.101 Observer 15.675b Observer * Type 1.650b Observer * Truedept 10.009 Observer * Type * Truedept 2.183 Resolution * Observer 3.094b Resolution * Observer * Type 1.405b Resolution * Observer * Truedept 2.104 Resolution * Observer * Type * Truedept 1.559 a
F ratio's are for Wilk's l; signi®cant at the .05 level
b
.000 .373 .000 .399 .000 .021 .001 .348 .000 .033 .000 .000 .000 .068 .000 .000
Exact statistic; cValues in bold are
Scanning resolution A Janhom et al
169
Figure 2 The eect of lesion depth on observer performance. This shows the eect of lesion depth for all three variable (signed, absolute observer error and observer con®dence) of all resolutions combined
Figure 3 The eect of resolution on observers for all depths combined. There were signi®cant dierences between 150 and 300 d.p.i. for absolute observer error and between 150 d.p.i. and the two higher resolutions for observer con®dence. For signed observer error, all three levels of resolution were signi®cantly dierent
observers only report being certain about the deepest lesions. The resolution main eect in the univariate analysis was signi®cant for signed, absolute observer error and con®dence (P=0.000 ± 0.010) whereas the lesion depth by resolution interaction eect was signi®cant only for absolute observer error (P=0.012). Figure 3 shows the resolution main eect for all depths combined. Signed observer error decreases significantly with increasing resolution. A signi®cant improvement in the scores for absolute observer error occurred only when the resolution was increased from 150 to 300 d.p.i. Increasing resolution from 300 to 600 d.p.i. showed no signi®cant dierence in lesion detectability. Observers were less con®dent about their observation at 150 d.p.i. than 300 or 600 d.p.i. The dierence in their con®dence was not found between viewing images scanned at 300 or 600 d.p.i. The eect of resolution at dierent lesion depths on absolute observer error is presented in Figure 4. It shows that the eect of resolution on observing dentinal lesions was more dramatic than for other lesion depths. Three hundred d.p.i. images resulted in the lowest absolute observer error for almost all lesion depths except for DEJ lesions. Additional testing of the eects of resolution and lesion depth was done and the results are shown in Table 2. In this analysis, the eect of resolution and lesion depth for the variable absolute observer error was signi®cant for sound surfaces and dentinal lesions. Increasing the resolution from 150 to 300 d.p.i. did not aect the absolute observer error for sound surfaces, but a further increase to 600 d.p.i. resulted in a larger absolute observer error. Viewing of dentinal lesions at 150 d.p.i. in comparison to 300 or 600 d.p.i. resulted in a higher absolute observer error.
Table 2 The resolution effect at different lesion depths for absolute observer error. Significant differences were found in the images of sound surfaces and dentinal lesions Lesion Absolute error
none outer enamel enamel + dej dentine
F
Sig.b
5.908a .445a 1.565a 4.428a
.003 .642 .212 .013
a
Exact statistic; bValues in bold are signi®cant at the .05 level
There was no signi®cant dierence between 300 and 600 d.p.i. in absolute observer error when viewing dentinal lesions. All eects mentioned so far are to some degree observer dependent, because all eects involving the observer in the multivariate analysis were signi®cant. The observer eect is also signi®cant in the univariate level. In Figure 5, the observer and resolution interaction eect is plotted for the signed observer error. Increasing the resolution makes hardly any dierence for seven out of 10 observers. Observer number eight is aected most by the change of resolution. The test was repeated without the scores of observer number eight to check whether the results were in¯uenced by this observer, but they remained the same. The test of the interaction eects of resolution and observer performance showed that for about half of the observers changing the resolution did not significantly aect their scores. The signi®cant dierence between types of element (premolar and molar) was found only on a personal level since only the interaction eect between element type and observer were signi®cant. Dentomaxillofacial Radiology
Scanning resolution A Janhom et al
170
Figure 4 The eect of resolution at dierent lesion depths on absolute observer error. This shows the interaction of resolution and lesion depth. For the sound surface, 600 d.p.i. was worse than 150 and 300 d.p.i. but for dentinal lesions 150 d.p.i. was the worst. Three hundred d.p.i. images resulted in the lowest absolute observer error for almost all lesion depths except for DEJ lesions
Figure 5 The eect of resolution on signed observer error for all 10 observers is plotted. Three observers, number four, three and eight, were more aected by the dierence of resolutions than the rest. Observer number eight was the one who was aected most by the change of resolution
Discussion The purpose of this study was to determine a proper scanning resolution for the digitization of bitewing radiographs on a ¯atbed scanner. Special emphasis was placed on the detection of approximal caries as a task. We compared varying resolutions (150, 300 and 600 d.p.i.) when diagnosing approximal caries. We also took into account the depth of the lesion. The Dentomaxillofacial Radiology
signed observer error, absolute observer error and the observer con®dence were variables calculated to measure relevant aspects of lesion detection. These measures were calculated in such a way that they could illustrate the amount of over- or under-estimation of lesion depth, the competence of the observers in detecting caries and the strength of their con®dence. We used analysis of variance to determine whether the resolution or any other factors signi®cantly in¯uenced the measures described above. It appeared that changing the resolution aected assessment of lesion depth, since the resolution main eect was signi®cant. It should be noted, however, that this might not be true as a general conclusion for all observers and for all depths. This is because the interaction eect between observer, resolution and lesion depth was also signi®cant. The result from univariate tests showed that the interaction between lesion depth and resolution was signi®cant for the absolute observer error for sound surfaces and dentinal lesions. For sound surfaces, the absolute observer error was larger when the observers viewed 600 d.p.i. images compared with 150 and 300 d.p.i. But for dentinal lesions, the highest error was found at 150 d.p.i. and there was no signi®cant dierence between 300 and 600 d.p.i. The interaction eect of resolution and observer also showed that there was no signi®cant dierence in about half of the observers. That means that, although on average the resolution eect is signi®cant, many observers were not aected by the change of resolution. The eect was also limited to certain lesion depths. The observer main eect was signi®cant. Some observers are more sensitive to the change in resolution than others. This means that a dierent group of observers would also aect the ®nal results. However, when we left out the results of one particular observer who scored very dierently from the others, the results were the same. The degree of experience of the observers in detecting caries on a monitor may have in¯uenced the results here. Some of the observers recruited in this study may not have been as familiar as others viewing images on a screen. The same may be true for their experience in diagnosing caries. We, however, did not ®nd a clear dierence between the radiologists and the cariologists. The gamma setting (of 2.2) for the scanner in this study was somewhat subjectively selected as providing good quality images. A dierent setting would probably have aected all resolutions in a similar manner. More research is required to investigate this assumption. The size of the images on the screen may also have aected our ®ndings. The 150 d.p.i. image is quite small on a 17 inch monitor, while the 600 d.p.i. occupies the full screen and the 300 d.p.i. about onequarter of the screen. Mùystad et al. found that digital image magni®cation has a signi®cant in¯uence on observer performance in the detection of approximal caries.12 They stated that the response rate obtained
Scanning resolution A Janhom et al
from 618 and 630 magni®ed images was signi®cantly inferior to that of 63, 66 and 612 magni®cations. In our study, 600 d.p.i. with a full screen image showed the highest absolute observer error on sound surfaces compared with the two lower resolutions. Where caries is concerned, however, 600 d.p.i. performed as well as 300 d.p.i. It is known that when the pixel size decreases, the noise level relative to the signal increases.13 This may, therefore, have caused the larger error in diagnosing sound surfaces. It is also possible that the real size of the image as shown on the monitor could be a more important factor than the actual scanning resolution. Versteeg et al. reported that there was no signi®cant dierence in locating the tip of an endodontic ®le between the original and zoomed-in half-size images (with the same size as the original but containing less information).14 Further investigation is needed to explain the eect of the size of the image. A direct comparison could not be made between this study and others that have used ROC curves as a means of analysis. However, we can relate the results from the present study to other studies by comparing the pixel size of the various image modalities. Ohki et al. concluded that digitization with a pixel size of 100 mm is adequate for the detection of caries.6 If we convert the resolution of 150, 300 and 600 d.p.i. in our study into pixels, the corresponding pixel size is approximately 170, 85 and 40 mm, respectively. Both
300 d.p.i. (85 mm) and 600 d.p.i. (40 mm) are smaller than the minimum size recommended by Ohki et al., which could explain the almost similar performance we found for 300 and 600 d.p.i. images. In summary, scanning bitewing radiographs with a ¯atbed scanner at 300 d.p.i. provides the best results for caries detection at almost all lesion depths. The absolute observer error from using a resolution of 150 d.p.i. was signi®cantly higher than that with 300 or 600 d.p.i. for dentinal lesions. This may have importance in the clinical situation. The diagnostic con®dence increases as the resolution increases but no signi®cant dierence was shown between resolutions of 300 and 600 d.p.i. The clinical bene®t, therefore, of 600 d.p.i. when compared with 300 d.p.i. is questionable. This is also important, because at 300 d.p.i., the digital ®le size is smaller and therefore the storage requirements lower. We conclude that in order to keep the ®le size at a minimum without losing the information necessary for caries diagnosis, scanning bitewing radiographs with a ¯atbed scanner at 300 d.p.i. is sucient.
171
Acknowledgement The authors would like to thank all ten observers who participated in this study. We thank Dr Phil Mileman for editorial assistance.
References 1. Pysher L, Harlow C. Teleradiology using low-cost consumeroriented computer hardware and software. AJR Am J Roentgenol 1999; 172: 1181 ± 1184. 2. Berry RF, Barry MH. Evaluation of a personal-computer-based teleradiology system serving an isolated Canadian community. Can Assoc Radiol J 1998; 49: 7 ± 11. 3. Kassebaum DK, McDavid WD, Dove SB, Waggener RG. Spatial resolution requirements for digitizing dental radiographs. Oral Surg Oral Med Oral Pathol 1989; 67: 760 ± 769. 4. Shrout MK, Potter BJ, Yurgalavage HM, Hildebolt CF, Vannier MW. 35-mm ®lm scanner as an intraoral dental radiographic digitizer I: A quantitative evaluation. Oral Surg Oral Med Oral Pathol 1993; 76: 502 ± 509. 5. Shrout MK, Potter BJ, Yurgalavage HM, Hildebolt CF, Vannier MW. 35-mm ®lm scanner as an intraoral dental radiographic digitizer II: Eects of brightness and contrast adjustments. Oral Surg Oral Med Oral Pathol 1993; 76: 510 ± 518. 6. Ohki M, Okano T, Nakamura T. Factors determining the diagnostic accuracy of digitized conventional intraoral radiographs. Dentomaxillofac Radiol 1994; 23: 77 ± 82. 7. Yuasa H, Ariji Y, Ohki M, Naitoh M, Shiojima M, Ushida M, Ariji E. Joint Photographic Experts Group compression of intraoral radiographs for image transmission on the World Wild Web. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1999; 88: 93 ± 99.
8. Jones DEA. A note on back-scatter and depth doses for elongated rectangular X-ray ®elds. Br J Radiol 1948; 22: 342 ± 345. 9. Jones DEA, Raine HC, Mix D. Correspondence to the editor. Br J Radiol 1949, 22: 549 ± 550. 10. Svenson B, Peterson A. In¯uence of dierent developing solutions and developing times on radiographic caries diagnosis. Dentomaxillofac Radiol 1990; 19: 157 ± 160. 11. Thylstrup A, Fejerskov O. Clinical and pathological features of dental caries. In: Thylstrup A, Fejerskov O, (eds). Textbook of clinical cariology. Copenhagen: Munksgaard textbook, 1994, pp 111 ± 157. 12. Mùystad A, Svanñs DB, Larheim TA, GroÈndahl H-G. Eect of image magni®cation of digitized bitewing radiographs on approximal caries detection: an in vitro study. Dentomaxillofac Radiol 1995; 24: 255 ± 259. 13. Hildebolt CF, Bartlett TQ, Brunsden BS, Hente NL, Gravier MJ, Walkup RK, et al. Bitewing-based alveolar bone densitometry: digital imaging resolution requirements. Dentomaxillofac Radiol 1994; 23: 129 ± 134. 14. Versteeg CH, Sanderink GCH, Lobach SR, Van der Stelt PF. Reduction in size of digital images: does it lead to loss detectability or loss of diagnostic information? Dentomaxillofac Radiol 1998; 27: 93 ± 96.
Dentomaxillofacial Radiology
