Prediction of osteoporosis with dental radiographs and age

Transcription

1 (2009) 38, The British Institute of Radiology RESEARCH Prediction of osteoporosis with dental radiographs and age JGC Verheij*,1, WGM Geraets 1, PF van der Stelt 1, K Horner 2, C Lindh 3, K Nicopoulou-Karayianni 4, R Jacobs 5, EJ Marjanovic 6, JE Adams 6 and H Devlin 2 1 Department of Oral and Maxillofacial Radiology, Academic Centre for Dentistry Amsterdam, Amsterdam, The Netherlands; 2 School of Dentistry, University of Manchester, Manchester, UK; 3 Department of Oral Radiology, Malmö University, Malmö, Sweden; 4 Department of Oral Diagnosis and Radiology, University of Athens, Athens, Greece; 5 Oral Imaging Centre, University of Leuven, Leuven, Belgium; 6 Department of Imaging Science and Biomedical Engineering, University of Manchester, Manchester, UK Objectives: In this study age and the trabecular pattern present on dental radiographs were used to predict the presence of osteoporosis. The objective was to evaluate the contribution of the trabecular pattern to the prediction. Methods: In this project, 671 women between 45 and 71 years of age were recruited. Medical history was obtained and dental radiographs were made. Bone mineral density (BMD) was measured at three sites to assess the presence of osteoporosis according to the World Health Organization criteria. The radiographs were subjected to image analysis methods yielding measurements of the trabecular pattern. Thereafter, discriminant analysis was used to predict the presence of osteoporosis by means of the trabecular pattern and age. Sensitivity and specificity of age and the trabecular pattern were compared. Also, it was checked whether the inclusion of the trabecular pattern improved the sensitivity and specificity that were obtained when only age was used as the predictor. Results: The sensitivity and specificity of the trabecular pattern present on dental radiographs were almost equal to those of age. However, combining age with the trabecular pattern increased the sensitivity from 0.71 to 0.75 and the specificity from 0.72 to 0.78; the latter increase was statistically significant. Conclusions: The trabecular pattern predicts the presence of osteoporosis just as well as age does. When combining the trabecular pattern with age, the sensitivity and specificity increased. Only the latter increase was statistically significant. (2009) 38, doi: /dmfr/ Keywords: osteoporosis; sensitivity; specificity; trabecular pattern Introduction *Correspondence to: Dr JGC Verheij, Department of Oral and Maxillofacial Radiology, Academic Centre for Dentistry Amsterdam, Louwesweg 1, 1066 EA Amsterdam The Netherlands; H.Verheij@acta.nl Received 13 June 2008; revised 23 September 2008; accepted 6 October 2008 According to the definition of the World Health Organization (WHO) formulated in 1994, osteoporosis is a systemic disease characterized by low bone mineral density (BMD), deterioration of bone structure and increased bone fragility. From the various methods available to measure BMD, at present the preferred technique is dual X-ray absorptiometry (DXA) at the hip or lumbar spine or both. It is known that after the age of 35 the BMD of men and women gradually decreases with increasing age. 1,2 Women tend to lose BMD more rapidly than men, especially after the menopause. As a result, osteoporosis is three times more common among women than men. In white women the risk of fractures of the spine, hip or wrist after the age of 50 years due to osteoporosis is estimated at 40 50%, similar to the risk of coronary heart disease. 3,4 To detect early signs of osteoporosis and to select those women who can benefit most from therapy would require BMD testing of all post-menopausal women at risk. 5 Because many women are involved this would require extensive facilities and high costs. Moreover, this would again require the time of both the women and the medical personnel. Therefore, there is a need for less expensive alternative methods of assessing the skeletal status that can be used on a large scale. Dental radiographs are relatively inexpensive and are already used in a large part of the adult population. Because osteoporosis is a systemic disease involving all skeletal bones and because

2 432 JGC Verheij et al dental radiographs always depict some mandibular or maxillary bone, dental radiographs may offer an opportunity for a screening tool for osteoporosis. Should this approach turn out to be feasible, then the general dental practitioner might fulfil the same role with respect to osteoporosis as for example to the early detection of oral cancer and refer the patient to a facility for bone densitometry when osteoporosis is suspected. 6 Characteristics of the trabecular pattern have been shown to have a fairly high precision and correlate reasonably well with BMD. 7,8 The objective of the current study was to predict the presence of osteoporosis using the trabecular pattern present on dental radiographs. To gain insight in the possible benefits, the prediction using the trabecular pattern was compared with the prediction using age. This study concentrates on sensitivity and specificity because they measure the quality of a diagnostic instrument. In clinical practice the diagnostic tool should increase the diagnostic accuracy, and this is the case when sensitivity and specificity are high. Materials and methods In 2003 the European Union provided a grant for a research project in five European Universities: Amsterdam, Athens, Leuven, Malmö and Manchester. This project, named Osteodent, investigated the diagnostic validity of dental radiography for identifying osteoporosis. The overall aim of the research project was to find methods which dentists can use to assess osteoporosis by means of dental radiographs. Subjects Within the project, 671 women in the age range years were recruited in the clinics of Athens, Leuven, Malmö and Manchester. The subjects included women attending for routine or emergency dental care, women working in the localities of the universities involved and others who had heard about the study by word of mouth or through local press articles. The project was approved by the ethical committees of the universities involved. The women s osteoporotic status was determined by measuring BMD at the lumbar spine and left hip. A dental panoramic radiograph and two intraoral radiographs were made. The medical history was recorded to exclude women with possible secondary osteoporosis caused by primary hyperparathyroidism, poorly controlled thyrotoxicosis, malabsorption, liver disease or alcoholism. Of the recruited women, 607 had a complete set of BMD values and dental radiographs and these women were used in the statistical analysis. BMD and the gold standard The BMD was measured by DXA scans of the lumbar spine (L1 L4) and left hip (total hip and femoral neck). The scans were made using the Hologic QDR 4500, the Hologic Discovery (Hologic, Bedford, MA) and the GE Lunar Prodigy (GE Lunar Corporation, Madison, WI) at the Athens, Leuven, Malmö and Manchester clinics. Shewarts rules were used to monitor quality assurance throughout the study period. 9 The European spine phantom and the method described by Pearson and colleagues 10 were used to standardize the DXA measurements of the different equipment used. This eliminates differences in equipment and makes the DXA measurements comparable. Next, they were converted into T-scores using the Hologic reference data for the lumbar spine and the NHANES (National Health and Nutrition Examination Survey) reference data for the left femoral neck and total left hip. 11 A T-score indicates how many standard deviations the DXA measurement of a subject differs from the mean DXA of a reference population of young women. Osteoporosis was determined according to the criteria from the WHO. According to these criteria a subject is classified as having osteoporosis if the T-score of the lumbar spine, left femoral neck or total left hip is 22.5 or less. This diagnosis is considered as the gold standard. Radiographs and regions of interest From each subject one panoramic and two intraoral radiographs were made. The panoramic radiographs were made with a Planmeca Promax device (64 66 kv) (Planmeca Oy, Helsinki, Finland), a Planmeca PM2002CC (66 68 kv) (Planmeca Oy), a Soredex Cranex Tome (70 kv) (Soredex, Helsinki, Finland) or a Soredex Cranex 3+ (69 kv) (Soredex). The Soredex Cranex Tome used photostimulable phosphor plates (ADC Solo, Mortsel, Belgium) which were scanned with a resolution of 79 pixels cm 1 (200 pixels inch 1 ). The other panoramic devices used conventional film cassettes which were scanned with a resolution of 252 pixels cm 1 (641 pixels inch 1 ), which was lowered to 79 pixels cm 1 (200 pixels inch 1 ) before making measurements. The intraoral radiographs were made with three Planmeca Prostyle Intra devices (60 63 kv) (Planmeca Oy), and one Siemens Heliodent MD (60 kv) (Sirona, Bensheim, Germany). They depicted the upper right and lower right premolar region on conventional films, which were scanned at a resolution of 118 pixels cm 1 (300 pixels inch 1 ). Four regions of interest containing mandibular or maxillary bone were selected manually, two on the panoramic radiograph and one on each intraoral radiograph. On the panoramic radiograph, the regions of interest were chosen in the right half of the lower jaw, one below the molars and the premolars, and the other in the ramus (Figure 1a). On average, the size of the region of interest was pixels in the front and pixels in the ramus. On the intraoral radiographs, the region of interest was chosen by preference between the roots of a premolar and a molar. Inclusion of parts of the adjacent roots was permitted (Figure 1b,c) because in a previous study it

3 JGC Verheij et al 433 Figure 1 Location of regions of interest on dental radiographs. (a) Panoramic radiograph; (b) intraoral radiograph of the lower jaw; and (c) intra-oral radiograph of the upper jaw had been shown that inclusion of parts of the adjacent roots does not influence the correlation of the trabecular pattern with BMD. 12 On average, the size of the region of interest on the intraoral radiograph was pixels for the lower jaw and pixels for the upper jaw. Measurements All measurements on the dental radiographs were carried out on the complete region of interest. The mean and standard deviation were computed from the unfiltered grey values in the region of interest. Next, low- and high-frequency information was removed from the region of interest and thereafter the region was segmented to obtain a binary version, consisting of black and white segments. The procedures to binarize the region of interest have been described previously The binarized region of interest was used to measure the number of black segments, number of white segments, total area of black segments, total area of white segments, total perimeter of white segments and fractal dimension. The measurements of number of segments, area and perimeter were standardized by dividing them by the total area of the region of interest. Then the white segments in the binary version of the region of interest were eroded into a wire frame that was used to measure the total length of struts, number of endpoints and the number of furcations. Similarly, the black segments in the binary version of the region of interest were eroded into a wire frame that was used to measure the total length of struts, number of endpoints and the number of furcations. The measurements made on the wire frame structure were standardized by dividing them by the total area of the region of interest. Finally, the binarized region of interest was used to measure the line frequency deviation (LFD) along 12 directions starting with 0, and then in steps of 15 up to 165. The method of measuring LFD has been described previously The LFD is intended to quantify spatial anisotropy. All the predictors were measured in each of the four regions of interest and are summarized in Table 1. Because age is an important determinant of osteoporosis that can be assessed easily without radiation and without cost, it is also included as a predictor in this study.

4 434 JGC Verheij et al Table 1 Description of characteristics of the trabecular pattern of each region of interest Predictors from the unfiltered grey values 1 Mean of the grey values in the region of interest 2 Standard deviation of the grey values in the region of interest Predictors from the black and white segments 3 Number of black segments * 4 Number of white segments * 5 Area of black segments * 6 Area of white segments * 7 Perimeter of white segments * 8 Fractal dimension Predictors from the white wire frame 9 Length of struts * 10 Number of endpoints * 11 Number of furcations * Predictors from the black wire frame 12 Length of struts * 13 Number of endpoints * 14 Number of furcations * LFD index of orientation of black and white segments 15 LFD orientation along 0 16 LFD orientation along LFD orientation along LFD orientation along LFD orientation along LFD orientation along LFD orientation along LFD orientation along LFD orientation along LFD orientation along LFD orientation along LFD orientation along 165 * Variables are standardized through division by the area of the region of interest Statistics The gold standard based on the DXA measurements was used to identify the subjects who were osteoporotic and those who were not. Discriminant analysis was used to find the variables that significantly contribute to the prediction of the presence of osteoporosis. 22,23 In discriminant analysis a latent variable is assumed that is a linear combination of the predictors. A standardized weight or relative importance for each predictor is determined in such a way that the means of both groups on the latent variable are separated most, i.e. that sensitivity and specificity attain their highest value. After the weights are obtained, a subject is classified as osteoporotic or not using only the significant predictors and the likelihood ratio principle. Differences in sensitivity and specificity were tested using the McNemar x 2 test. 24 All statistical computations were done using the SPSS package. Results The mean age of the 607 women included in the statistical analysis was 54.8 years, with a standard deviation of 6.1. Frequencies of osteoporotic and nonosteoporotic women are presented in Table 2, which clearly shows an increasing percentage of osteoporotic women as age increases. Seven predictors were found Table 2 Percentage of osteoporotic and non-osteoporotic women within age groups according to the gold standard. The 95% confidence intervals and frequencies are given in parentheses Age (years) Osteoporosis No osteoporosis Total % (3 11%; 9) 93% (89 97%; 122) % (5 14%; 18) 91% (87 95%; 174) % (21 35%; 40) 72% (65 79%; 103) % (23 42%; 30) 68% (58 77%; 63) % (41 73%; 21) 43% (29 57%; 16) % (16 75%; 5) 55% (25 84%; 6) 11 Total 20% (17 23%; 123) 80% (77 83%; 484) 607 that were significantly related to the presence of osteoporosis (x , degrees of freedom (df) 5 7, P ); the canonical correlation coefficient was These predictors are presented in Table 3 together with their standardized weight (relative importance) and the correlation with the latent variable. In line with previous research, age was the most important predictor. The other six predictors are from the dental radiographs. Sensitivity and specificity are presented in Table 4 and the receiver operating characteristic (ROC) curves for the three prediction methods in Figure 2. The best prediction is obtained when age is combined with the characteristics of the trabecular pattern. Compared with age, specificity increases significantly from 0.72 to 0.78 (x , df 5 1, P ). Sensitivity increases from 0.71 to 0.75, but this was not statistically significant. Using only age or only the trabecular pattern leads to less accurate predictions, and their sensitivity and specificity are not significantly different. Discussion The first conclusion of this study is that characteristics of the trabecular pattern on dental radiographs can predict the presence of osteoporosis with sensitivity 0.70 and specificity 0.69, which is just as good as the prediction using age. Next it is concluded that combining age with the trabecular pattern gives a statistically significant increase in specificity. This implies that analysis of the trabecular pattern on dental radiographs helps to prevent referring healthy women unnecessarily to a facility for measuring BMD. For determining the specificity, 484 observations were available but for determining the sensitivity only 123; therefore, the McNemar test for the significance of the sensitivity had less power than for testing the specificity. Therefore, it is likely that a significantly increased sensitivity will be found in a study with a larger number of osteoporotic subjects. In line with the results of previous research, age is the most important predictor. 1,2,8 Because bone strength decreases with age, the correlation between bone strength and age is negative. Similarly, the correlation between the latent variable and age is negative. Therefore, we may speculate that the variable can be

5 JGC Verheij et al 435 Table 3 Predictors for osteoporosis, their standardized weight (relative importance) and the correlation with the latent variable. All predictors are significant at a~0:05 Predictor Weight Correlation Age Panoramic, front, white wire frame, number of endpoints Intraoral, lower jaw, white segments, area Intraoral, upper jaw, LFD Panoramic, front, LFD Intraoral, lower jaw, LFD Intraoral, lower jaw, mean grey value interpreted as resistance against osteoporosis. This is also supported by the positive correlation of the latent variable with BMD, i.e with the BMD of the total hip, 0.45 with the BMD of the femoral neck and 0.49 with the BMD of the lumbar spine. Because the six predictors of the trabecular pattern are from the panoramic as well as from intraoral radiographs, it was not clear whether the panoramic radiographs predict osteoporosis better than the intraoral radiographs. However, five of the six variables measure bone characteristics from the lower jaw. Therefore, radiographs from the lower jaw seem to predict osteoporosis better than radiographs from the upper jaw. Characteristics of the trabecular pattern are clearly related to osteoporosis and have a reasonable sensitivity and specificity by themselves. However, there is no clear explanation for the underlying mechanism. The results from the discriminant analysis may give some indication of this mechanism. As noted above, the latent variable that is used in this analysis may be interpreted as resistance against osteoporosis. Table 3 further confirms this interpretation because of the positive correlation with the mean grey value, which may be interpreted as a rough measure of bone density because denser objects give brighter projections. The area of white segments and the number of endpoints in the white skeleton are less easy to interpret because they are computed after a highpass filter is applied. The area of white regions is positively correlated with the latent variable, and thus it may be concluded that when this area increases bone strength also increases. The number of endpoints in the white skeleton is negatively correlated with the latent variable, which means that bone strength decreases when the number of endpoints increases. This result is in line with the perforation or discontinuation hypothesis of Mosekilde 25 and Parfitt, 26 which states that osteoporosis is accompanied by perforation of the trabecular structure, and thus leads to more endpoints in the skeleton of the trabecular structure. However, the reverse relation has also been reported. 20,27 The identification of the underlying mechanism is hampered by the nature of radiographic imaging. The presence of soft tissue and the projection of a three-dimensional (3D) object on to a two-dimensional (2D) surface are two inherent problems. From the clinical point of view the results may not appear interesting because it is simpler to determine the age of a patient than to analyse the radiographic trabecular pattern of the patient. However, the present research is part of a series of studies that try to find information that is contained in the trabecular pattern and how the pattern changes during growth and ageing. This study shows that the trabecular pattern of women aged contains at least as much information on the BMD as age, which is the main determinant of BMD. Further research might find image features of the trabecular pattern that increase the predictive power of the trabecular pattern. Moreover, it would be interesting to study the prediction of fractures. With respect to hip fractures, age is an even stronger predictor than BMD. 28 Furthermore, it is conceivable that, in the future, the trabecular pattern on dental radiographs could be analysed on a routine basis, requiring only an assistant to define a region of interest and a common office computer to do the analysis and give a warning in case a suspect pattern is recognized. The success of a prediction model depends not only on the right choice of predictors and the noise present in their measurements but also on the noise present in the criterion variable: the BMD in the present study. When noise is present in the criterion variable, a 100% correct prediction is not possible. It is known that DXA measurements have accuracy errors ranging from 3% to 15%, so a more technical line of research may be to improve the BMD measurement. 29,30 Acknowledgment This work was supported by a research and technological development project grant from the European Commission Fifth Framework Programme Quality of Life and Management of Living Resources (QLK ; Osteodent ). Table 4 Sensitivity and specificity of age and trabecular pattern. The 95% confidence intervals are given in parentheses Gold standard Prediction No osteoporosis Osteoporosis Specificity False positive No osteoporosis Age + trabecular pattern 0.78 ( ) 0.22 ( ) Age 0.72 ( ) 0.28 ( ) Trabecular pattern 0.69 ( ) 0.31 ( ) False negative Sensitivity Osteoporosis Age + trabecular pattern 0.25 ( ) 0.75 ( ) Age 0.29 ( ) 0.71 ( ) Trabecular pattern 0.30 ( ) 0.70 ( )

6 436 JGC Verheij et al Sensitivity Specificity Figure 2 Receiver operating characteristic (ROC) curves for the prediction of osteoporosis. x, prediction using age; m, prediction using trabecular pattern; &, prediction using age and trabecular, optimal sensitivity and specificity pattern;n References 1. Riggs BL, Wahner HW, Dunn WL, Mazess RB, Offord KP, Melton LJ III. Differential changes in bone mineral density of the appendicular and axial skeleton with aging. J Clin Invest 1981; 67: Riggs BL, Wahner HW, Seeman E, Offord KP, Dunn WL, Mazess RB, et al. Changes in bone mineral density of the proximal femur and spine with aging. J Clin Invest 1982; 70: Chrischilles EA, Butler CD, Davis CS, Wallace RB. A model of lifetime osteoporosis impact. Arch Intern Med 1991; 151: Melton LJ III, Chrischilles EA, Cooper C, Lane AW, Riggs BL. Perspective. How many women have osteoporosis? J Bone Miner Res 1992; 7: National Osteoporosis Foundation. Review of the evidence for prevention, diagnosis and treatment and cost-effective analysis. Osteoporos Int 1998; 4(Suppl.): S7 S Sciubba JJ. Oral cancer and its detection. History taking and the diagnostic phase of management. J Am Dent Assoc 2001; 132: Geraets WGM, Verheij JGC, Van der Stelt PF, Horner K, Lindh C, Nicopoulou-Karayianni K, et al. Osteoporosis and the general dental practitioner: reliability of some digital dental radiological measures. Community Dent Oral Epidemiol 2007; 35: Geraets WGM, Verheij JGC, Van der Stelt PF, Horner K, Lindh C, Nicopoulou-Karayianni K, et al. Prediction of bone mineral density with dental radiographs. Bone 2007; 40: Orwoll ES, Oviatt SK. Longitudinal precision of dual-energy X- ray absorptiometry in a multicenter study. The Nafarelin/Bone Study Group. J Bone Miner Res 1991; 6: Pearson J, Dequeker J, Henley M, Bright J, Reeve J, Kalender W, et al. (1995) European semi-anthropomorphic spine phantom for the calibration of bone densitometers: assessment of precision, stability and accuracy. The European Quantitation of Osteoporosis Study Group. Osteoporos Int 1995; 5: Looker A, Wahner H, Dunn W, Calvo M, Harris T, Heyse S, et al. Updated data on proximal femur bone mineral levels of US adults. Osteoporos Int 1998; 8: Geraets WGM, Verheij JGC, Van der Stelt PF, Horner K, Lindh C, Nicopoulou-Karayianni K, et al. Selection of region of interest on intraoral radiographs for the prediction of bone mineral density. Dentomaxillofac Radiol 2008; 37: Geraets WGM, Van der Stelt PF, Netelenbos CJ, Elders PJM. A new method for automatic recognition of the radiographic trabecular pattern. J Bone Miner Res 1990; 5: Geraets WGM, Van der Stelt PF. Analysis of the radiographic trabecular pattern. Pattern Recogn Lett 1991; 12: Geraets WGM, Van der Stelt PF, Elders PJM. The radiographic trabecular bone pattern during menopause. Bone 1993; 14: Korstjens CM, Geraets WGM, Van Ginkel FC, Prahl-Andersen B, Van der Stelt PF, Burger EH. Longitudinal analysis of radiographic trabecular pattern by image processing. Bone 1995; 17: Korstjens CM, Mosekilde L, Spruijt RJ, Geraets WGM, Van der Stelt PF. Relations between radiographic trabecular pattern and biomechanical characteristics of human vertebrae. Acta Radiol 1996; 37: Korstjens CM, Geraets WGM, Van Ginkel FC, Prahl-Andersen B, Van der Stelt PF, Burger EH. An analysis of the orientation of the radiographic trabecular pattern in the distal radius of children. Growth Develop Aging 1994; 58: Geraets WGM, Van der Stelt PF, Lips P, Elders PJM, Van Ginkel FC, Burger EH. Orientation of the trabecular pattern of

7 JGC Verheij et al 437 the distal radius around the menopause. J Biomech 1997; 30: Geraets WGM, Van der Stelt PF, Lips P, Van Ginkel FC. The radiographic trabecular pattern of hips in patients with hip fractures and in elderly control subjects. Bone 1998; 22: Geraets WGM. Comparison of two methods for measuring orientation. Bone 1998; 3: Morrison DF. Multivariate statistical methods. London: McGraw-Hill, Finn JD. Multivariance: univariate and multivariate analysis of variance, covariance, regression and repeated measures. Chicago: National Educational Resources, Hays WL. Statistics. New York: Holt, Rinehard and Winston, Mosekilde L. Vertebral bone quality and strength, SPINE: State of the Art Reviews 1994; Vol 8, 1: Parfitt AM. Trabecular bone architecture in the pathogenesis and prevention of fracture. Am J Med 1987; 82: White CW, Rudolph DJ. Alterations of the trabecular pattern of the jaws in patients with osteoporosis. Oral Surg Oral Med Oral Pathol Oral Radiol Endod 1999; 88: Hui SL, Slemenda CW, Johnston CC. Age and bone mass as predictors of fracture in a prospective study. J Clin Invest 1988; 81: Baran DT, Faulkner KG, Genant HK, Miller PD, Pacifici R. Diagnosis and management of osteoporosis: guidelines for the utilization of bone densitometry. Calcif Tissue Int 1997; 61: Genant HK, Engelke K, Fuerst T, Glüer CC, Grampp S, Harris ST, et al. Noninvasive assessment of bone mineral and structure: state of the art. J Bone Miner Res 1996; 11: