KuUendorff B, Petersson K, Rohlin M. Direct digital radiography for the detection of periapical bone lesions: a clinical study. Endod. Dent Traumatol 1997; 13: ...
Endod Dent Traumatol 1997; 13: 183-189 Printed in Denmark . All rigiits reserved
Copyright © Munksgaard
1997
Endodontics & Dental Traumatology ISSN 0109-2502
Direct digital radiography for the detection of periapical bone lesions: a clinical study KuUendorff B, Petersson K, Rohlin M. Direct digital radiography for the detection of periapical bone lesions: a clinical study. Endod Dent Traumatol 1997; 13: 183-189. © Munksgaard, 1997. Abstract — The aim was to compare the observer performance of direct digital radiography, with and without image processing, with that of conventional radiography, for the detection of periapical bone lesions. Eor 50 patients, a conventional periapical radiograph using E-speed film was taken. Then, a direct digital image of the same area was made. The images presenting the periapical bone tissue of 59 roots were assessed by seven observers using a 5-point confidence scale. The digital images were first presented as original images, with default contrast and brightness set by the computer system. Following this, the observers were allowed to use the processing facilities for greyscale treatment. The results for original and processed direct digital images and for conventional radiographs were compared by Receiver Operating Characteristic (ROC) analysis. The area under the ROC curve, calculated as P(A) value, was 0.88 for conventional film, 0.82 for original digital images and 0.78 for processed images. Corresponding A^ values were slightly higher, 0.89, 0.84 and 0.81. Statistically significant differences between ROC areas calculated as P(A) values for the methods were found. Comparison between A^ values showed no significant differences between conventional radiographs and original digital images, whereas the difference between A^ values for original and processed digital images was still significant. It was concluded that conventional film radiography performed slightly better for the detection of periapical bone lesions than direct digital radiography and that image processing did not improve the observer performance.
A periapical bone lesion visualised by radiography is the only sign of asymptomatic apical periodontitis. It is therefore important that new radiographic techniques are evaluated regarding the diagnosis of periapical bone tissue. The development of direct digital radiographic methods has made it possible to reduce the radiation dose and to enhance the image quality after image acquisition. The diagnostic accuracy of direct digital radiography has been shown to be comparable to that of conventional film radiography for the detection of experimental bone tissue lesions. In a study on periodontal bone lesions, Eurkart et al. (1)
B. Kuiiendorff, K. Petersson^, iVi. Rohiin^ Departments of ^Oral Radiology and ^Endodontics, Centre for Oral Health Sciences, Lund University, Malmo, Sweden
Key words; digital radiography; image processing; periapicai periodontitis; radiographic image enhancement; radiography, dentai Boel KuUendorff, Department of Oral Radiology, Centre for Orai Health Sciences, Carl Gustafs vag 34, 21421 Maimo, Sweden Accepted March 3, 1997
found that the artificial bone lesions were detected as well with the Sens-A-Ray system as on film. Yokota et al. (2) found direct digital images (RVG system) to be superior to conventional films for the detection of periapical lesions, while conventional radiography was significantly better when no lesion existed. Similar results were presented by Tirell et al. (3), who found direct digital radiography (RVG system) to outperform film for the detection of the smallest lesions in a study of chemically created, cortical bone lesions, while no difference was found for baseline images or larger lesions. Eor the detection of periapical bone 183
Kuiiendorff et ai. lesions in dry human mandibles, the diagnostic accuracy of direct digital images (Visualix system) was shown to be comparable to that of conventional film images (4). Only a few clinical studies on direct digital radiography of the periapical bone tissue have been presented. Farman et al. (5) found that Visualix/VIXA2 images, processed with greyscale equalisation, were superior to conventional radiographs for measuring the dimensions of periapical bone lesions in patients undergoing periapical surgery. A comparison between E-speed film and direct digital radiography, using a storage phosphor-plate system (Digora®), showed that there was no difference in diagnostic accuracy between the imaging systems for the detection of periapical lesions in vivo (6). An advantage of digital radiography is the possibility of changing the image mathematically with the aid of different image processing functions, such as alteration of contrast and brightness and filtering of images. In this way, an image can be enhanced in several ways, according to the diagnostic task and the observer's requirements. Wenzel & Hintze (7) reported that dentists subjectively preferred the processed image to the original version. In another study on observers' use of image enhancement facilities, it was found that all observers used image processing in almost all digital images (8). In the dental literature, the impact of image processing on observer performance has only recently been reported. Farman et al. (5) reported that processed Visualix images were superior to the original images for the measurement of periapical bone defects. For the detection of experimental periapical lesions, image processing was found to improve the performance of some observers, but the overall result was not significantly improved (9). The most useful processing functions were increase of contrast and alteration of brightness. The aims of this clinical study were: • To investigate whether direct digital radiography was comparable with conventional film radiography for the detection of periapical bone lesions and • To determine the impact of image processing on the detection of periapical bone lesions. iViateriai and methods Equipment The X-ray source (Oralix, Gendex Dental Systems, Milan, Italy) was operated at 65 kV and 10 mA with a total filtration of 2 mm Al equivalent. The distance between focus and the rectangular coUimator opening was 20 cm. Conventional film radiographic images were taken with
E-speed film (Kodak Ektaspeed, Eastman-Kodak Co, Rochester, PsTY, USA). Films were processed in an 184
automatic processor (Periomat, Diirr Dental, Beitigham, Germany) with a processing time of 7 min. The films were mounted in opaque frames (Trollhatteplast, Trollhattan, Sweden). Direct digital radiographic images were produced with a
Visualix/VIXA system (Gendex Dental Systems, Milan, Italy) provided with a charge-coupled device (CCD) sensor of 18.1 mmX24.2 mm active area. The image matrix was 288X384 with a pixel size of 63 \imX63 jam. Images were displayed on a monitor (Ultra VGA, Eite-on, Taipei, Taiwan) operated with a screen resolution of 1024X768 pixels. The sensor used in this study was of the second generation (VIXA-2), equipped with a scintillation layer. Patient examination Fifty patients were examined. None of the patients presented with subjective symptoms. Forty patients were attending a recall examination 1 to 4 years after endodontic treatment and 10 patients were referred to the Department of Oral Radiology for full-mouth radiographic examination. The patients were informed about the aims of the study by the first author (BK). First, a conventional radiograph was taken with the aid of a filmholder by another examiner. Then, with the patient's permission, an additional image with direct digital radiography (DDR) was made by the first author. Permission from the local ethics committee to perform an additional X-ray exposure with DDR had been obtained. The direct digital image was made with the sensor shielded by a disposable plastic cover and placed in a specially designed filmholder (Hawe-Neos, Switzerland) during exposure. The exposure time was decreased by 40% compared with the exposure time set for E-speed film. About 20% of the total images had to be retaken, mainly because of problems in depicting the periapical area. The direct digital image was stored in the original form. The examined teeth were selected so as to represent all tooth groups and an appropriate proportion of teeth with and without periapical bone lesions (Table 1). Each root was counted as a separate object for assessment. Table 1. Distribution of roots with and without a radiographic periapical bone lesion which were assessed by the observers Incisors
Cuspids Premolars
Molars
Total
Upper jaw Lesion No iesion
3 6
2 5
2 4
2 11
g 26
Lower jaw Lesion No lesion
0 3
2 4
2 7
1 5
5 19
12
13
15
19
Total
59
Direct digitai radiography of periapicai hone iesions Interpretation of images Initially, one observer, an oral radiologist with 10 years' clinical experience of oral radiology, assessed all conventional radiographs. This observer had previously participated in an in vitro study on periapical lesions (4, 9) and presented a diagnostic accuracy, expressed as an ROC value, P(A), of 0.92 fbr conventional radiography. For each tooth, this observer made a decision based on a 5-point rating scale: 1. Definitely no lesion 2. Probably no lesion 3. Uncertain 4. Probably a lesion 5. Definitely a lesion. Independently, the procedure was repeated by the first author. Decisions that both observers agreed on served as the "true radiographic diagnosis". Images where these observers were uncertain (3 on the rating scale) were excluded from the study. This was the case for 14 of the initial 73 roots. For the remaining 59 roots, the expert observers decided that a lesion was present for 14 roots and no lesion for 45 roots. The images of the 59 roots were presented to seven observers: three oral radiologists, two endodontists and two general dental practitioners undergoing further training in oral radiology. They were asked to assess the periapical bone tissue of the indicated tooth, using the same 5-point rating scale as described above. First, the observers assessed all conventional film radiographs using a lightbox and a Mattsson magnifier (DAB Dental, Sweden). Two weeks later, they were presented with the direct digital images. The digital images were displayed on the monitor of the Visualix system, in a room with subdued light. The observer-screen distance was 50-60 cm. First, the observers assessed the original image, with default contrast and brightness set by the computer system. After this, they were allowed to use different image processing facilities for greyscale treatment provided by the software of the system (10). Then, a second decision was made. In all, three decisions were recorded for each observer and each root. Image processing The Visualix software functions include functions for greyscale treatment, image enhancement filtering and zooming. In this study, only the greyscale treatment functions were used. Equalisation (E) activates an algorithm for greyscale equalisation. The grey level spectrum of the image is extended in order to utilise the entire greyscale range of the system. This function is also known as histogram equalisation. The actual improvement in image quality may vary, depending on the settings of the original image. Less level/more level decreases or increases the number of displayed
grey levels by a factor of 2. The maximum and default number is 64. Contrast (C+, C-) increases or decreases image contrast by fixed steps, thus changing the slope of the density diagram of the point in the image (pixels). Brightness (B+, B—) increases or decreases image brightness by shifting the video output level in fixed steps of 10%, thus changing the pixel value uniformly. Analysis of observations The ratings of the seven observers for E-speed film, DDR original images and DDR processed images were compared with the "true radiographic diagnosis" and the diagnostic performance of each observer and image form was measured as a Receiver Operating Characteristic (ROC) value. The areas under the ROC cur\'es were calculated as individual and mean P(A) values and as A^ values using the ROCFIT computer program for likelihood estimation (11). The P(A) value is the area under an ROC curve where points representing the true positive rate, TPR, (sensiti\dt\') and false positive rate, FPR (1 - specificity) are plotted on linear probability scales. The A,^ value represents the area under a straight ROC graph, plotted on a binormal scale. The ROC values of each image form were statistically compared pairwise, using Wilcoxon's signed rank test, and all three methods together, using Friedman's test. A value of p
