Diagnosis of osteoporosis from dental panoramic radiographs using the support vector machine method in a computer-aided system
© Kavitha et al; licensee BioMed Central Ltd. 2012
Received: 25 August 2011
Accepted: 16 January 2012
Published: 16 January 2012
Early diagnosis of osteoporosis can potentially decrease the risk of fractures and improve the quality of life. Detection of thin inferior cortices of the mandible on dental panoramic radiographs could be useful for identifying postmenopausal women with low bone mineral density (BMD) or osteoporosis. The aim of our study was to assess the diagnostic efficacy of using kernel-based support vector machine (SVM) learning regarding the cortical width of the mandible on dental panoramic radiographs to identify postmenopausal women with low BMD.
We employed our newly adopted SVM method for continuous measurement of the cortical width of the mandible on dental panoramic radiographs to identify women with low BMD or osteoporosis. The original X-ray image was enhanced, cortical boundaries were determined, distances among the upper and lower boundaries were evaluated and discrimination was performed by a radial basis function. We evaluated the diagnostic efficacy of this newly developed method for identifying women with low BMD (BMD T-score of -1.0 or less) at the lumbar spine and femoral neck in 100 postmenopausal women (≥50 years old) with no previous diagnosis of osteoporosis. Sixty women were used for system training, and 40 were used in testing.
The sensitivity and specificity using RBF kernel-SVM method for identifying women with low BMD were 90.9% [95% confidence interval (CI), 85.3-96.5] and 83.8% (95% CI, 76.6-91.0), respectively at the lumbar spine and 90.0% (95% CI, 84.1-95.9) and 69.1% (95% CI, 60.1-78.6), respectively at the femoral neck. The sensitivity and specificity for identifying women with low BMD at either the lumbar spine or femoral neck were 90.6% (95% CI, 92.0-100) and 80.9% (95% CI, 71.0-86.9), respectively.
Our results suggest that the newly developed system with the SVM method would be useful for identifying postmenopausal women with low skeletal BMD.
Osteoporosis is a disease that develops asymptomatically in its early stages and is characterized by low bone mass and micro-architectural deterioration of bone tissue, thereby increasing the risk of fractures . The incidence is higher in developed countries, primarily because they have a large elderly population. It is a major health problem in the Japanese elderly population as well and is estimated to affect approximately 12 million people . A huge number of dental panoramic radiographs, offering greater opportunities for studying bones, are taken every year . The features of osteoporosis can often be observed in the films of affected individuals, and there are significant relationships between the mandibular cortical bone quality, quantity and bone mineral density (BMD) [4, 5]. Panoramic radiographs were previously used  for developing a mandibular cortical index to assess the porosity of cortical bone. Many studies have shown a correlation between the mandibular cortical width (MCW) on dental panoramic radiographs and BMD at the hip , lumbar spine  and forearm , the most common sites of fracture related to osteoporosis in postmenopausal women. The costs associated with advanced imaging techniques and distribution of the equipment limit their accessibility for large segments of populations and broad-based screening examinations.
Recently, computer software-assisted diagnostic techniques have been used because they reduce the influence of subjective human interpretation and have often increased diagnostic accuracy . Osteoporosis screening using a computer-aided diagnosis (CAD) system developed in previous studies needed manual assistance in measuring the MCW [11, 12]. Furthermore, MCW measured at only one point (below the mental foramen) may influence measurement error. However, the image of the hyoid bone sometimes overlaps the cortex below the mental foramen on dental panoramic radiographs, which can result in measurement errors with CAD systems such as the one used in the previous study . Continuous measurement of the MCW between the upper and lower boundary at each point below the mental foramen could help to overcome this problem in order to accurately detect osteoporosis .
Most research effort on the analysis of orthopaedic X-ray images has been focused on the detection of osteoporosis using methods of texture and fractal analysis [14–16]. Continuous measurement by CAD systems utilizes a trimmed mean method to achieve good diagnostic results. However, using a trimmed mean requires prior threshold setting . Unfortunately, the selection of an initial parameter setting for the prior threshold may affect results drastically. Support vector machine (SVM) methods have the feasibility and superior ability to extract higher-order statistics and have become extremely popular for classification and prediction. As a continuation of our extensive previous work of developing a CAD system , we decided to apply the CAD system with SVM-kernel methods to diagnose and classify women with low BMD. This study employs a newly adopted kernel-based radial basis function (RBF)-SVM method instead of a trimmed mean method to improve its diagnostic efficacy and achieve a higher accurate classification rate for the discriminating and identifying women with low BMD from those with normal skeletal BMD. The application of the CAD system used in this study was directed towards problems in classification. The goal of classification was to accurately determine to which set (or class) an unknown item belongs.
The Hiroshima University Human Subject Committee approved the study protocol, and dental panoramic radiographs were taken for all the subjects after informed consent was obtained. A total of 531 women underwent skeletal BMD examinations at an oral radiology clinic at Hiroshima University Hospital between 1996 and 2001. This study included 100 postmenopausal women as subjects, of whom 60 were allocated to the training of the tool and 40 to its testing. All 100 women underwent BMD of the lumbar spine (L2-L4) and femoral neck by dual-energy X-ray absorptiometry (DPX-alpha; Lunar Co., Madison, WI, USA). The inclusion criteria were postmenopausal women aged 50 years or older with no previous diagnosis of osteoporosis. The exclusion criteria were subjects who menstruated less than a year, had any metabolic bone disease (hyperparathyroidism, hypoparathyroidism, Paget's disease, osteomalacia, renal osteodystrophy or osteogenesis imperfecta), had cancer with bone metastasis, had significant renal impairment, had history of taking medication known to affect bone metabolism (e.g. oestrogen), had undergone hysterectomy or oophorectomy, had a history of smoking, had any bone destructive lesion (e.g. malignant tumours or osteomyelitis) in the mandible or had any spinal fracture. Spinal fractures were confirmed semi-quantitatively on lateral radiographs.
The subjects were classified as normal (T-score > -1.0), osteopenia (T-score between -1 and -2.5) or osteoporotic (T-score < -2.5) at each skeletal site according to the World Health Organization (WHO)  criteria. The Adult Health Study cohort in Japan  reported that the cut-off BMD value of osteoporosis in the lumbar spine that was based on the Japanese definition  (less than 70%) was similar to that based on the WHO definition (T-score < -2.5 SD); therefore, we used the WHO definition in this study.
Dental panoramic radiography
All of the panoramic radiographs were obtained by using an AZ-3000 Panoramic Dental X-ray (Asahi Co., Kyoto, Japan) at 12 mA and 15 s; kVp values varied between 70 and 80, and a flat-bed scanner (ES-8000; Epson, Tokyo, Japan) was used to digitize the images at a resolution of 300 dpi. Screens of speed group 200 (HG-M; Fuji Photo Film Co., Tokyo, Japan) and film (UR-2; Fuji Photo Film Co.) were used. One set of duplicate films (MI-Dup; Fuji Photo Film Co.) that were processed with an automatic film processor (Cepros M; Fuji Photo Film Co.) comprised the 100 original panoramic radiographs for the assessment. The appearance of the mandibular inferior cortex was clear bilaterally in the radiographs.
Computerized scheme for automated detection of osteoporosis
Area of interest determination
The boundary of each object was not sharp, so we removed all areas that were considered to be background and applied enhancement processes to the remaining objects. The typical histogram equalization method is the first step in image enhancement to obtain new enhanced images with a uniform histogram. The most common method of thresholding assigns a pixel to one class if its grey level is greater than a specified threshold and otherwise assigns it to the other class for separating objects from its background.
We applied an averaging filter as a low-pass filter  to the multiplied image and subtracted this low-frequency image from the multiplied image to obtain an image that contained only high frequencies. This step was applied to images that no longer had background illumination variations. Therefore, the high-pass filtering process sharpened the boundary along the cortical bone only because the region of no interest adjacent to the cortical bone had been removed.
Cortical margin determination
Two parameters are required to optimize the RBF kernel of the SVM classifier γ that determines the capacity of the RBF kernel and C, the regularization parameter. These parameters must be adjusted to obtain the best classification with a reduced number of support vectors because this number is directly related to the speed of execution.
We chose to use the RBF as the kernel function because it was shown to perform well on our datasets for classifications using average and variance results of the continuous measurements between women with low BMD (BMD T-score of - 1.0 or less) and normal skeletal BMD. Quadratic programming was used to optimize the combination of parameters and found to yield better classification results at γ = 1 and C = 1. Smaller values were used to avoid reproducing noise and avoid over-fitting to the data samples that were used in the training procedure. The RBF parameter and weighting factors were determined by experimentation on the training samples. We employed a data analysis framework written in Matlab, which incorporates freely available SVM tools for Matlab that were implemented by Scholkopf , to perform classification. The mean of the MCW on both sides of the mandible was used in this study. The risk-index range that corresponded to a sensitivity of approximately 90% was chosen to determine the optimal cut-off threshold. The sensitivities, specificities, positive predictive values (PPV), negative predictive values (NPV), accuracies and likelihood ratios for the positive (LR+) results for identifying subjects with low BMD were calculated. The accuracy of the classifications of 10 randomly selected subjects for dental panoramic radiographs measured twice with a one-month interval by the same examiner also was evaluated.
Results and Discussion
Performance of the proposed SVM method for identifying women with low lumbar spine BMD and femoral neck BMD at a 95% confidence interval (CI)
Positive predictive value %
Negative predictive value %
Likelihood ratio (+) %
Performance of the proposed SVM method for identifying women with combined skeletal bone mineral densities (BMD) at a 95% confidence interval (CI)
Positive predictive value %
Negative predictive value %
Likelihood ratio (+) %
We used our proposed SVM method with a CAD system and dental panoramic radiographs to diagnose women with low BMD easily and quickly. The use of the SVM kernel in this study provided a high degree of consistency and reproducibility in the results. One of the key advantages of this CAD system over a manual assessment is the objectivity of the automated evaluation. The proposed CAD system directly assesses the bones on radiographs by measuring the MCW continuously between the mental foramen and mandibular angle, which can reduce measurement errors that occur with the conventional assessment . The trimmed mean method (accuracy = 79%) of the recently proposed CAD method  can be replaced by the SVM method (accuracy = 88%) to obtain a better result. The proposed SVM diagnostic model performs differential diagnosis very well. Because classification by the trimmed mean method requires prior threshold setting, the SVM approach for classifying women as having either a low or normal BMD may be a better choice.
In the previously developed CAD systems [11, 12], the diagnostic accuracy of the radiologist was adversely affected by manual interactions. Arifin et al.  reported that measurement of the MCW at one point (below the mental foramen) by CAD had a detection sensitivity of 88% and specificity of 58.7%. The conventional method of continuous measurements using the trimmed mean technique  was reported to have a sensitivity of 90% and specificity of 75% on the basis of the lumbar spine BMD and a sensitivity of 81.8% and specificity of 69.2% on the basis of the femoral neck BMD. However, for our newly proposed SVM method, the sensitivity and specificity were 90.9% and 83.8%, respectively on the basis of the lumbar spine BMD and 90.0% and 69.6%, respectively on the basis of the femoral neck BMD.
These findings indicate that the classification performance of the diagnostic system using the SVM method achieved higher accuracy for detecting women with low BMD compared with the conventional approach. The difference between these results is reasonable because unlike other techniques, the CAD system with SVM has an advantage of converging the problem to the global optimum and not to a local optimum.
Our findings are supported by those of previous studies. Lim et al.  evaluated bone fractures from X-ray images with different classifiers and reported a high classification accuracy of 98.2% from a method using a combination of SVM classifiers. Caligiuri et al.  showed that their method was promising for discriminating between healthy and fractured bones with high Az values. It was also reported that the sensitivity and specificity for an RBF kernel-SVM that was used for the reorganization of nuclear receptors was almost equal to our classification system . Comparisons of our experimental results with those of previous studies demonstrated the feasibility and excellent performance of our proposed system in diagnosing high-risk groups with low BMD or osteoporosis. Several screening tools based on simple questionnaires have been developed to identify postmenopausal women with low skeletal BMD or osteoporosis, and validation of these tools has also been performed in many countries . The sensitivity and specificity of such decision rules in identifying postmenopausal women with osteoporosis ranged from 90% to 92% and 37%-45%, respectively. This proves that the diagnostic efficacy of our SVM method is better than that of the several questionnaire-based screening tools that were used in the previous studies, although the backgrounds of the subjects were different. The limitations in our study were that the number of subjects was relatively small, and the subjects were relatively healthy postmenopausal women because we used rigid exclusion criteria.
The diagnostic efficacies achieved by the application of an RBF kernel-SVM in our study showed that our CAD system was effective and accurate for identifying women with low BMD. Compared with the previously developed trimmed mean method, the SVM method was shown to be more accurate (88% vs. 79%) and could classify women as having either low or normal BMD more efficiently. On the basis of the highly satisfactory sensitivity and specificity results, the proposed system is expected to be a helpful tool for classifying women with low BMD and can provide a second diagnosis that may reduce misdiagnoses. Our SVM method is considered to be a reliable choice for the proposed system because it is fast and specific for classification, using dental panoramic radiographs, of postmenopausal women with low BMD.
MSK is thankful to the Graduate School of Engineering, Hiroshima University, for providing financial support during her study period (2009-2012). This work was supported in part by a Grant-in-Aid from the Japan Society for the Promotion of Science to AT (No. 21592404).
- National Institute of Health: Osteoporosis prevention, diagnosis and therapy. NIH Consensus Statement. 2000, 17: 1-45.Google Scholar
- Muraki S, Yoshimura N: Incidence of and prognosis for osteoporotic fracture. Clin Calcium. 2006, 16: 1431-7. (in Japanese)PubMedGoogle Scholar
- Taguchi A: Triage screening for osteoporosis in dental clinics using panoramic radiographs-A Review. Oral Dis. 2010, 16: 316-27. 10.1111/j.1601-0825.2009.01615.x.View ArticlePubMedGoogle Scholar
- Horner K, Devlin H: The relationship between mandibular bone mineral density and panoramic radiographic measurements. J Dent. 1998, 26: 337-43. 10.1016/S0300-5712(97)00020-1.View ArticlePubMedGoogle Scholar
- Horner K, Devlin H: The relationships between two indices of mandibular bone quality and bone mineral density measured by dual energy X-ray absorptiometry. Dentomaxillofac Radiol. 1998, 27: 17-21. 10.1038/sj.dmfr.4600307.View ArticlePubMedGoogle Scholar
- Klemetti E, Kolmakov S, Kröger H: Pantomography in assessment of the osteoporosis risk group. Scand J Dent Res. 1994, 102: 68-72.PubMedGoogle Scholar
- White SC, Taguchi A, Kao D, Wu S, Service SK, Yoon D, et al: Clinical and panoramic predictors of femur bone mineral density. Osteoporos Int. 2005, 16: 339-46. 10.1007/s00198-004-1692-4.View ArticlePubMedGoogle Scholar
- Taguchi A, Tanimoto K, Suei Y, Ohama K, Wada T: Relationship between the mandibular and lumbar vertebral bone mineral density at different postmenopausal stages. Dentomaxillofac Radiol. 1996, 25: 130-35.View ArticlePubMedGoogle Scholar
- Devlin H, Horner K: Mandibular radiomorphometric indices in the diagnosis of reduced skeletal bone mineral density. Osteoporos Int. 2002, 13: 373-78. 10.1007/s001980200042.View ArticlePubMedGoogle Scholar
- Devlin H, Horner K: Diagnosis of osteoporosis in oral health care. J Oral Rehab. 2008, 35: 152-57.Google Scholar
- Arifin AZ, Asano A, Taguchi A, Nakamoto T, Ohtsuka M, Tsuda M, Kudo Y, Tanimoto K: Computer-aided system for measuring the mandibular cortical width on dental panoramic radiographs in identifying postmenopausal women with low bone mineral density. Osteoporos Int. 2006, 17: 753-9. 10.1007/s00198-005-0045-2.View ArticlePubMedGoogle Scholar
- Devlin H, Allen PD, Graham J, Jacobs R, Karayianni K, Lindh C, Van der Stelt PF, Harrison E, Adams JE, Pavitt S, Horner K: Automated osteoporosis risk assessment by dentist: A new pathway to diagnosis. Bone. 2007, 40: 835-42. 10.1016/j.bone.2006.10.024.View ArticlePubMedGoogle Scholar
- Kavitha MS, Samopa F, Asano A, Taguchi A, Sanada M: Computer-aided measurement of mandibular cortical width on dental panoramic radiographs for identifying osteoporosis. J Inv Clin Dent. 2011, 2: 1-9. 10.1111/j.2041-1626.2011.00051.x.View ArticleGoogle Scholar
- Geraets WG, der Stelt PV, Elders PJ: The radiographic trabecular bone pattern during menopause. Bone. 1993, 14: 859-64. 10.1016/8756-3282(93)90315-2.View ArticlePubMedGoogle Scholar
- Link T, Majumdar S, Konermann W, Meier N, Lin J, Newitt D, Ouyang X, Peters P, Genant H: Texture analysis of direct magnification radiographs of vertebral specimens: Correlation with bone mineral density and biomechanical properties. Acad Radiol. 1997, 4: 167-76. 10.1016/S1076-6332(05)80286-7.View ArticlePubMedGoogle Scholar
- Ouyang X, Majumdar S, Link TM, Lu P, Lin JC, Newitt DC, Genant HK: Morphometric texture analysis of spinal trabecular bone structure assessed using orthogonal radiographic projections. Med Physi. 1998, 25: 2037-45. 10.1118/1.598391.View ArticleGoogle Scholar
- World Health Organization: Assessment of fracture risk and its application to screening for postmenopausal women osteoporosis. 1994, Geneva: WHOGoogle Scholar
- Fujiwara S, Kasagi F, Masunari N, Naito K, Suzuki G, Fukunaga M: Fracture prediction from bone mineral density in Japanese men and women. J Bone Miner Res. 2003, 18: 1547-53. 10.1359/jbmr.2003.18.8.1547.View ArticlePubMedGoogle Scholar
- Orimo H, Hayashi Y, Fukunaga M, Sone T, Fujiwara S, Shiraki M, Kushida K, et al: Diagnostic criteria for primary osteoporosis: year 2000 revision. J Bone Miner Metab. 2001, 19: 331-37. 10.1007/s007740170001.View ArticlePubMedGoogle Scholar
- Gonzalez R, Woods R: Digital image processing. 1992, Addison-Wesley publishing company, 3Google Scholar
- Kavitha MS, Li L, Samopa F, Asano A, Taguchi A: Continuous measurement of mandibular cortical bone in dental panoramic radiographs for the diagnosis of osteoporosis using a clustering algorithm on histograms. Proc Sec APSIPA ASC. 2010, 560-67.Google Scholar
- Samopa F: Tooth shape measurement on dental radiographs for forensic personal identification. Chapter 5, Ph.D. Dissertation. 2009, Information Engineering, Hiroshima University, 2-56.Google Scholar
- The nature of statistical learning theory. 1995, Springer-Verlag, Berlin Heidelberg, New YorkView ArticleGoogle Scholar
- Vapnik VN: Statistical Learning Theory. 1998, Wiley, New-YorkGoogle Scholar
- Wee LJK, Tan TW, Ranganathan S: SVM-based prediction of caspase substrate cleavage sites. BMC Bioinformatics 7 Suppl. 2006, 5: 14-View ArticleGoogle Scholar
- Burges C: A tutorial on support vector machines for pattern recognition. Data Min Know Disc. 1998, 2: 121-67. 10.1023/A:1009715923555.View ArticleGoogle Scholar
- Scholkopf B, Smola AJ: Learning with Kernels. 2002, MIT Press, Cambridge, MAGoogle Scholar
- Lim SE, Xing Y, Chen Y, Leow WK, Howe TS, Png MA: Detection of femur and radius fractures in X-ray images. Proc 2nd Int Conf on Advances in Medical Signal and Information Processing. 2004, 249-56.Google Scholar
- Caligiuri P, Giger ML, Favus M: Multifractal radiographic analysis of osteoporosis. Med Physi. 1994, 21: 503-8.View ArticleGoogle Scholar
- Cai J, Li Y: Classification of nuclear reactors subfamilies with RBF kernel in support vector machine. 2005, Springer-Verlag, Berlin, Heidelberg, 680-85.Google Scholar
- Cadarette SM, McIsaac WJ, Hawker GA, Jaakkimainen L, Culbert A, Zarifa G, Ola E, Jaglal SB: The validity of decision rules for selecting women with primary osteoporosis for bone mineral density testing. Osteoporos Int. 2004, 15: 361-6. 10.1007/s00198-003-1552-7.View ArticlePubMedGoogle Scholar
- The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1471-2342/12/1/prepub
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.