Radiographic detection of approximal caries: a ... - BIR Publications

ã

Dentomaxillofacial Radiology (2000) 29, 312 ± 318 2000 Macmillan Publishers Ltd. All rights reserved 0250 ± 832X/00 $15.00 www.nature.com/dmfr

Radiographic detection of approximal caries: a comparison of dental ®lms and digital imaging systems K Syriopoulos*,1, GCH Sanderink1, XL Velders1 and PF van der Stelt1 1

Department of Oral Radiology, Academic Centre for Dentistry Amsterdam (ACTA), The Netherlands

Objectives: To compare the diagnostic accuracy for the detection of approximal caries of two dental X-ray ®lms, two CCD-based digital systems and two storage phosphor (SP) digital systems. Methods: Fifty-six surfaces in 56 extracted unrestored premolars were radiographed under standardised conditions using two E-speed dental ®lm, Ektaspeed Plus (Eastman Kodak Co, Rochester, NY, USA) and Dentus M2 Comfort (Agfa-Gevaert, Mortsel, Belgium), two CCD systems, Sidexis (Sirona, Bensheim, Germany) and Visualix (Gendex, Milan, Italy) and two SP systems, Digora (Soredex, Helsinki, Finland) and DenOptix (Gendex, Milano, Italy). The images were assessed by eight observers (four radiologists and four general practitioners). True caries depth was determined by histological examination. True caries depth was subtracted from the values given by the observers and an analysis of variance was performed. The null hypothesis was rejected when P50.05. Results: No signi®cant dierences were found in diagnostic accuracy with the two dental ®lms and the Sidexis and Digora systems. The depth of the lesion signi®cantly aected observer performance. Caries depth was underestimated. Radiologists performed signi®cantly better than general practitioners whatever the recording system. Conclusion: The diagnostic accuracy of digital systems is comparable with that of dental ®lms. The ability of dentists to recognise cariers correctly is the main factor contributing to variation in radiographic diagnosis and not the imaging modality. Keywords: dental caries; radiography, dental; observer variation; diagnostic errors

Introduction The rapid advances in computer technology have had a signi®cant impact on dental radiography. In 1987 the ®rst direct digital system became commercially available as an alternative to conventional radiography. Since then several systems have been introduced on the market. There are two fundamentally dierent concepts for direct digital image acquisition, the CCD-based (charged-coupled device) and the storage phosphor (SP) systems. In the CCD system, a chip is used as a sensor for the radiation. A cable connects the sensor to the computer and the image is displayed almost immediately on the

*Correspondence to: K Syriopoulos, Department of Oral Radiology, Academic Centre for Dentistry Amsterdam (ACTA), Louwesweg 1, 1066 EA Amsterdam, The Netherlands Received 11 February 2000; accepted 24 May 2000

monitor after exposure. In the SP system, a phosphor plate is exposed and a latent image stored. The information contained in the plate is released by exposure to a laser scanner. The way the phosphor plates are placed in the mouth, the image size and the fact that there is no wire attached to the plate means that a SP system is logistically similar to ®lmbased imaging. The ®rst SP system in dental radiography, the Digora, was introduced in 1994. It provides two sizes of imaging plates comparable with size #0 and #2 ®lm. A single plate can be scanned in approximately 30 s. Recently, an add-on facility has become available enabling up to 20 plates to be scanned without the operator intervening. In 1997, the DenOptix system was introduced. The system has ®ve sizes of imaging plates which are mounted in a carousel which can hold up to 29 imaging plates for scanning.1

Approximal caries K Syriopoulos et al

The reduction in processing time is regarded as one of the main advantages of digital radiography. In comparison with conventional ®lm digital systems provide facilities for image manipulation, e.g. contrast and brightness adjustments, subtraction radiography, 3-D reconstruction and automated image analysis, as well as a reduction in patient dose. As far as diagnostic accuracy is concerned, digital and conventional radiography give comparable results for determing root canal length,2 ± 4 assessing bone defects5 ± 7 and detecting caries.8 ± 10 The aim of this study was to compare the diagnostic accuracy for detecting approximal caries using conventional radiography, CCD and SP digital systems. Materials and methods To examine the diagnostic accuracy for approximal caries detection 60 unrestored extracted human premolars were selected. The teeth were mounted in plaster of Paris as close as possible to each other in groups of ®ve. Three teeth in the middle of each block were used in the study with the teeth at either end creating natural contact points. In total, 20 plaster blocks were constructed. For each radiograph two blocks were used simulating a bitewing radiograph. The exposures were made with a Heliodent MD (Siemens, Bensheim, Germany) operating at 60 kVp and 7 mA, 1.5 mm Al equivalent ®ltration and a halfvalue layer of 1.9 mm Al. Reproducible radiographs were obtained by placing each plaster block in a special designed holder mounted on the X-ray machine. The focus-®lm distance was 30 cm and the beam size 364 cm. A 12-mm thick soft tissue equivalent material was placed between the cone end and the tooth blocks.11 The radiographs were taken using six dierent recording systems: two dental ®lms, two CCD and two SP digital systems as follows: Dental ®lms:

(a) Dentus M2 (Agfa-Gevaert, Mortsel, Belgium) (b) Ektaspeed Plus (Eastman Kodak Co, Rochester, NY, USA). CCD systems: (a) Visualix II (Gendex, Milano, Italy) (b) Sidexis (Sirona, Bensheim, Germany) SP systems: (a) Digora (Soredex, Helsinki, Finland) (b) DenOptix (Gendex, Milano, Italy) The images produced with the Gendex DenOptix digital system were scanned and viewed at 300 and 600 d.p.i. resolution. Therefore, seven dierent recording systems were compared in the present study. Their characteristics are shown in Table 1. The exposure time for the two ®lms was 0.32, chosen so that the optical density of the dentine at the most prominent part of the approximal surface was as close as possible to 0.9.11 Although the CCD sensors and SP

Table 1 Characteristics of the seven imaging modalities tested in the study Imaging modality Gendex Visualix II Sirona Sidexis Soredex Digora Gendex DenOptix (300 d.p.i.) Gendex DenOptix (600 d.p.i.) Agfa Dentus M2 Kodak Ektaspeed Plus

Type of Active area File size sensor (mm) (kB)

313

Image size (pixels)

CCD CCD SPP SPP

20630 18630 31641 31641

306 267 229 197

6846456 6606412 5606416 5336377

SPP

31641

783

10646752

Film Film

31641 31641

± ±

26 magnification

plates allow a wide range of exposure settings, the exposure time was set at 50% of that of the E-speed ®lms as it is often recommended by the manufacturers. The eect of variation in the exposure time was not a subject of the present study. The conventional ®lms were processed manually with Agfa Dentus chemicals (Heraeus Kulzer, Dormagen, Germany).11 The volume of the developing tank was l l and it was placed in a thermostatically controlled bath containing circulating water. The developing time was 5 min at 208C. The ®lms of each manufacturer were from the same batch. The processed radiographs were mounted in frames in such a way that the identi®cation of the ®lm types was completely covered. They were examined using a viewing box with constant light intensity and a 26-magnifying viewer (X-Produkter, MalmoÈ, Sweden). The digital images were displayed on a SVGA 17inch monitor screen (Digital, Digital Equipment Co, Maynard, MA, USA) with help of special application software (Emago Dental Software, ODS, Amsterdam The Netherlands). Only one image was displayed at the time. The observers were not informed of the type of digital system under examination. They were allowed to adjust the brightness and contrast of the images. Figure 1 shows the dierences in size and active area of the images made with the four digital systems. The conventional radiographs and the digital images were examined by eight dentists all of whom worked at the Academic Centre for Dentistry Amsterdam (ACTA). Four of them were radiologists at the Department of Oral Radiology and the other four were general practitioners. They were asked to detect caries in the right approximal surfaces but were not informed about the number of lesions. Only one tooth surface was assessed as measurements of disease from two surfaces in the same tooth may not be independent of each other.12 The following four-point scale was used: 0=no caries; 1=lesion restricted to the enamel; 2=lesion reaching the amelodentinal junction (ADJ); 3=lesion extending into the dentine. The diagnostic performance of each observer with the seven recording systems was compared with the histological diagnosis. The histological examination was performed with an 86-magnifying stereomicroscope (Model 355110, Wild, Heerbrugg, Switzerland) Dentomaxillofacial Radiology


314

Chicago, IL, USA). The main statistical test was analysis of variance. The values for depth of caries determined from the histological examination were subtracted from the values recorded by the eight observers. The observers' estimate yielded an error of estimation. For instance, if the histological examination revealed a sound surface (score 0) and the diagnosis of the observer was enamel lesion (score 1), subtraction resulted in an estimation error score of 1 (absolute error) or +1 (net bias). In a similar manner, the observer's diagnosis of an actual dentinal lesion (score 3) as an enamel lesion (score 1) yielded an estimation error score of 2 (absolute error) or 72 (net bias). Estimation error of 0 indicated that the observers' diagnosis was the same as the outcome of the histological examination. Net estimation bias shows the direction of the estimation error, i.e. if the observer tends to underestimate (negative value) or overestimate (positive value) the severity of the lesion. However as the example in Table 2 shows, an overestimation may cancel out an underestimation. This directional bias is avoided when using the absolute estimation error. The analysis of variance was performed with the scores for estimation error as the `dependent variable'. The lesion depth scores derived from the histological examination were entered as a `between subject factor'. Observer, imaging modality and observer profession (radiologist or general practitioner) were entered as `within subject factors'.

Figure 1 Relative size of the images obtained with the ®ve digital systems. (A) Visualix, 6846456 pixels. (B) Sidexis, 6606412 pixels. (C) Digora, 5606416 pixels. (D) DenOptix (300 d.p.i.), 5336377 pixels. (E) DenOptix (600 d.p.i.), 10646752 pixels. Image E is twice the size of image D

by two observers jointly (KS and a cariologist) and served as the validation method for depth of caries. During sectioning four surfaces were damaged and were therefore excluded from the material. The ®nal material consisted of 56 approximal surfaces: 14 sound, 11 with enamel caries, eight with caries reaching the ADJ and 23 with dentinal caries. Data analysis Data were analysed using the Statistical Package for the Social Sciences (SPSS 7.5 for Windows, SPSS, Dentomaxillofacial Radiology

The null hypotheses were: (a) There were no signi®cant dierences between the seven imaging modalities in detecting approximal caries. (b) All eight observers performed equally well in detecting approximal caries. (c) Radiologists did not dier signi®cantly from the general practitioners in their ability to detect caries. (d) The depth of caries did not signi®cantly aect the detection of the lesion. (e) There were no signi®cant interaction eects between lesion depth, imaging modality and observer. A null hypothesis was rejected when P50.05.

Table 2 values

Average estimation error using absolute or net estimation

Tooth surface A B C D Mean error

Actual lesion depth

Observer assessment

1 3 1 2

1 2 3 1

Absolute estimation error

Net estimation bias

0 1 2 1 1

0 ±1 +2 ±1 0


Results The results of the statistical analysis are summarised in Table 3. The main eects of the imaging modality, observer and observer profession were signi®cant as were the interaction eects between histological lesion depth by observer and observer profession. There was a signi®cant dierence in accuracy with the seven imaging modalities (P=0.026) (Figure 2). The dierence in accuracy (estimation error) between Sidexis (the ®rst-rated system) and DenOptix (300 d.p.i.) (the last-rated) was 0.17. The analysis revealed signi®cant dierences between the imaging modalities but it did not show which were signi®cant. Pair-wise comparisons were therefore carried out between them (Table 4). No signi®cant dierences were found between the ®rst four imaging modalities, Sidexis (CCD), Dentus M2 ®lm, Ektaspeed Plus ®lm and Digora (SP). The ®rst three systems were signi®cantly better than Visualix II. Accuracy was least with the DenOptix (SP-300 d.p.i.) system. Accuracy of the DenOptix improved at 600 d.p.i. but the dierence was not signi®cant (P=0.127). The imaging modality by lesion depth interaction eect was not signi®cant (P=0.594). The severity of the lesion had no eect on the performance of the imaging modality. Some modalities performed consistently better than others whatever the lesion depth. Signi®cant dierences were found in the performance of the eight observers (P=0.000). They arrived at

dierent results examining the same images but this dierence was speci®c for any particular imaging modality. It was found that some observers performed consistently better than others, whatever the imaging modality (the observer by imaging modality interaction eect was not signi®cant, P=0.444). On the other hand the performance of the observers was signi®cantly aected by the depth of the lesion (the observer by histological lesion depth interaction eect was significant, P=0.000). Radiologists performed signi®cantly better than general practitioners (P=0.000). Figure 3 shows the estimation error of the four radiologists and the four general practitioners with the seven imaging modalities. Radiologists performed consistently better than general practitioners whatever the imaging modality (the observer profession by imaging modality interaction eect was not signi®cant, P=0.150). The estimation error made by the general practitioners using the Ektaspeed Plus ®lm or the Sidexis system was similar to that made by the radiologists using the DenOptix (300 d.p.i.) or Visualix systems. Figure 4 shows the estimation error made by each group of observers at each depth of lesion. Radiologists and general practitioners made a similar overestimation of the number of sound surfaces. However, when caries was present, the diagnosis of the

315

Table 3 Factors affecting caries diagnosis using two dental film types and five digital systems (levels of significance, P