The diagnostic validity of dental radiography techniques for identifying osteoporotic patients (OSTEODENT)
Intraoral radiographs from the > 600 subjects in the project were used to assess the trabecular pattern. Radiographs from the upper and lower premolar region were used. Reference radiographs were obtained from three patients in the study. These patients were diagnosed as osteoporotic, osteopenic or normal, respectively, from DEXA measurements of the hip and lumbar spine. The reference radiographs were not used for assessments of the trabecular pattern. The trabecular pattern in the radiographs was assessed according to three classes describing the trabecular pattern as dense, heterogeneous or sparse. Five observers assessed all radiographs and three observers assessed 10% of the radiographs twice. Before the assessments started the observers met for discussion and calibration. Intraobserver agreement calculated as kappa index showed moderate or good agreement while interobserver agreement was fair. The sensitivity for five observers’ assessments of the intraoral radiographs with sparse trabecular pattern as indicative of osteoporosis at either hip or spine varied between 13.2 and 39.1. The specificity varied between 82.4 and 95.6. Assessment of sparse trabecular pattern on intraoral radiographs offered a combination of low sensitivity but high specificity for osteoporosis diagnosis. If it is assumed that high specificity is preferred for osteoporosis assessment by dentists, then this method may have potential for clinical use, although inter-observer variability may be a problem.
Radiographic images of the jaws show a typical pattern as a result of the attenuation of the radiation by the trabecular bone. This pattern is not equivalent to the three-dimensional architecture of the trabecular bone, because the pattern is a two dimensional projection of the three-dimensional bone structure. It has, however, characteristics that are directly depending on the trabecular structure and dimensions. Therefore, the (two-dimensional) radiographic trabecular pattern can provide information about changes of the (three-dimensional) trabecular dimensions (Geraets et al., 1993; Korstjens et al., 1995, 1996). The image analysis technique is based on segmentation of darker and lighter areas in the radiographic image and the detection of the central axis of the dark and the light regions. Subsequently, the area of dark and light regions is determined and the average width, as well as the length, number of intersections and endpoints and number of segments as shown in the image of the axis of the light and of the dark regions. Finally the polar distribution of the direction of the trabecular pattern is determined. This work package focussed on the development and testing of software for the analysis of the trabecular pattern on intraoral radiographs of premolar regions of the upper and lower jaw. The objective was to predict the severity of osteoporotic changes in these subjects. The BMD values (see WP 1) served as the reference (golden standard) for the osteoporotic condition of each subject. A special software environment was developed to automate the analysis of larger number of radiographic images. This made it possible to analyse large series (or the complete data set) for a variety of ROIs. The ideal ROI is a region, which shows the trabecular pattern without overprojection of other dental structures. Intra-oral radiographs have a limited size and therefore they do not fulfil this requirement always. For this reason, we investigated the effect of a small part of adjacent teeth being visible in the ROI. If the effect is negligible, a larger ROI is feasible, otherwise the selection of the ROI is critical and must be done very careful. The difference of the outcome of the parameters on a ROI with and without a small area of overlapping tooth structures was not significant. Because the procedure of analysing the trabecular features was automated and therefore easy to carry out once set up, it was decided to include both options in the final analysis. The predicted BMD values for spine and hip, based on the trabecular pattern analysis on intra-oral radiographs and age as parameters produced ROC values in the range between 0.77 and 0.82. This shows that the software for the analysis of the trabecular pattern for the prediction of BMD values performed very well and is a promising tool for the detection of osteoporosis.
A database has been developed of 671 women (age 45 - 70 years) from four centres in Europe with data on BMD at: - Lumbar spine (L2-L4). - Left femoral neck. - Left total hip. All data cross-calibrated using the European Spine Phantom, and subject to continued quality control throughout the study duration.
This result is a database of computer-generated mandibular cortical width measurements obtained from dental panoramic radiographs using newly developed software. Volunteer female subjects in the 45 to 70 year age band, recruited from four European centres, underwent a Dental Panoramic Radiographic examination and bone densitometry. Two ASM methods, one entirely automatic and one manually initialised, were used to derive measurements of mandibular cortical width. 661 subjects were examined (mean age 54.8y; sd = 6.19y), with 140 (21.2%) being classified as having osteoporosis. Using the manually initialized ASM method, Az values for the diagnosis of osteoporosis at any site and at the femoral neck alone were 0.818, (95% CI: 0.786 to 0.847) and 0.839, (95% CI: 0.809-0.866), respectively. Using the automatically initialized ASM method, The Az values for the diagnosis of osteoporosis at any site and at the femoral neck alone were 0.762 (95% CI: 0.728 - 0.794) and 0.809; (95% CI: 0.777 to 0.838), respectively. The difference in Az of the two methods was significant (p<0.001). Conclusions: ASM-based methods of mandibular cortical width measurement were diagnostically effective in diagnosis of osteoporosis. The manually initialised method, involving a small amount of user interaction, performed best. Further analysis is needed to establish the appropriate diagnostic threshold for clinical use. Potential applications: This dataset is available for use in determining the diagnostic validity of cortical thickness in diagnosis of osteoporosis: the principal purpose of the OSTEODENT study. It may however, have applications in another context. We have interest in looking at the prospective value of cortical thickness as a means of assessing 'bone quality' in dental implant planning. End-users of results: In the context of osteoporosis, the end-users are: 1. the OSTEODENT consortium, 2. Other future collaborative researchers (subject to mutual agreement of consortium), 3. Industrial collaborators interested in developing the software, 4. Dental practitioners (subject to having the software). Main benefits: The innovative aspect of this dataset is in the context of the entire OSTEODENT project. It is part of a robust, comprehensive database of information on a pan-European sample that is larger than any previous set. This dataset may have commercial value in acting as a reference population for subsequent work.
Database of digital dental x-rays to be used for the analysis of the trabecular pattern and the validation of software tools for the detection of osteoporosis using dental radiographs. The database of image data is accessible per patient, per site, per type of radiograph (intra-oral and panoramic) and clinical data of the patients are available.
Result description: 671 volunteer female subjects in the 45 to 70 year age band, recruited from four European centres, underwent a Dental Panoramic Radiographic examination. Five observers, all oral radiologists but of different experience, made manual measurements of width of the mandibular lower border cortex below the mental foramina bilaterally, and classification of the porosity of the mandibular cortex using the previously published Klemetti index. Data of 661 subjects (mean age 54.8y; sd = 6.19y) were available for analysis, with 140 (21.2%) being classified as having osteoporosis. Mean within-subject variance for the five observers was 0.126mm (sd = 0.355mm). Repeatability is the difference between two measurements made by any pair of observers for the same subject and was expected to be less than 0.983 mm for 95% of pairs of observations. Potential applications: This dataset is available for use in determining the diagnostic validity of cortical thickness or cortical porosity in diagnosis of osteoporosis: the principal purpose of the OSTEODENT study. It may however, have applications in another context. We have interest in looking at the prospective value of cortical thickness/porosity as a means of assessing 'bone quality' in dental implant planning. End-users of results: In the context of osteoporosis, the end-users are: 1. the OSTEODENT consortium, 2. Other future collaborative researchers (subject to mutual agreement of consortium), 3. Dental practitioners (subject to receiving appropriate training). Main benefits: The innovative aspect of this dataset is in the context of the entire OSTEODENT project. It is part of a robust, comprehensive database of information on a pan-European sample that is larger than any previous set. This dataset may have commercial value in acting as a reference population for subsequent work.
Result Description: This study provides a unique database for future work in this field. This is because the database is one of the world's largest and most detailed and includes people sampled from a wide area across Europe. We were able to recruit patients from Athens (Greece), Malmo (Sweden), Manchester (United Kingdom) and Leuven (Belgium). Our results were therefore able to be generalized across Europe. The database contains detailed data about the clinical risk factors for osteoporosis and can be used to calculate all existing clinical risk indices. Dissemination: We have mobilized opinion in the partner countries to introduce dentists to the concept that they may be able to play a role in diagnosing osteoporosis. This has been done by way of newspaper articles, scientific journals and lectures to dentists and lay people.
The tool to measure jaw bone density on intraoral radiographs, which was developed in a first phase of the Osteodent project (Result 1) was used to analyse the radiographs of the >600 subjects in the project. Upper and lower jaw bone was analysed. Furthermore, all measurements were done by 5 different observers, and one observer re-measured 10% of the radiographs. Inter- and intra-observer reliability was very high, showing no significant difference between measurements of the 5 observers, nor between the repeated measurements of one observer. With ROC-analysis, for all observers, areas under the curve were above 0.700, pointing to a fair accuracy of the method.
A densitometric tool for jaw bone to analyse intraoral radiographs was developed. It consisted of custom made bite blocks, including an aluminium wedge for reference. Also, software to analyse the radiographs taken with this tool was developed and optimised. Precision, accuracy and minimal detection threshold for the analyses of intraoral radiographs were assessed. Methods: Bone samples deriving from the premolar region of 47 human mandibles were selected for analysis. The samples were obtained from adult cadavers in the department of anatomy (Faculty of Medicine, KULeuven) with ethical approval. Digital radiography was performed on all bone samples. Direct volumetric measurements served as gold standard density values and allowed determination of accuracy. DXA scans were performed on all specimens. For all radiographs, density in mm Al eq was calculated using a custom-made software, Osteop. Precision and intra- and interobserver reliability of this method was assessed. The bone specimens were progressively decalcified. At standard time intervals the percentage of decalcification was calculated. At each decalcification step, radiographs were taken and analyzed. All radiographs were also analysed, not using the wedge, but the average pixelvalue of the background-colour as a reference. This was done as to be able to detect the advantage of including the wedge. Results: CV was always lower then 3%, which points to a good precision of the method. Correlation between the density measurements in mm Al eq and the DXA results was .9, for the density measurements in mm Al eq and the direct density measurements r was .5. The custom-made software was able to detect a change in bone mineralization of 6.6%. The densitometric measurements with only the average pixelvalue of the background-colour as a reference, were not significantly correlated with the DXA-results. Conclusions: The present method for bone densitometric analysis offers potentials for clinical evaluation of bone density and minute bone density changes in the jaw bone. It was proven to be a superior method, compared to purely analysing pixel values. The densitometric tool was found to be a valid tool to measure bone density on intraoral radiographs. Therefore it was suitable to perform measurements on the large sample of intraoral radiographs collected in the Osteodent project.
This result consists of automatic image analysis software that can robustly and accurately measure the width of the inferior mandibular cortex on dental panoramic radiographs. These measurements can provide a useful indicator of osteoporosis and reduced bone mineral density (BMD). The width of the mandibular cortex can be measured automatically by computer and if this information is combined with a little clinical information, the end result is a powerful test for osteoporosis. The correlation between the cortical width and BMD was found to be highly significant. Receiver Operating Characteristic (ROC) analysis showed that clinically useful specificity and sensitivity results could be achieved. Potential applications for this result: This software has enormous potential for incorporation into commercially available dental digital imaging software used by dentists or, possibly, as a 'stand-alone' software add-on. It also has potential for incorporation into existing dental office management systems. Its diagnostic validity was high and involved little or no user interaction. End-users of results: Dental imaging equipment manufacturers. Dental office management systems. Software developers. Dentists as the ultimate end-user. Main benefits: Automatic computer assessment of osteoporosis by measuring the mandibular cortical thickness without inter or intra observer variability. Minimal effort by users. Good diagnostic validity.