The diagnosis of periodontal disease is greatly facilitated

MODERN RADIOGRAPHIC METHODS IN THE DIAGNOSIS OF PERIODONTAL DISEASE P.F. VAN DER STELT Department of Oral Radiology, Academic Center for Dentistry Amsterdam (ACTA), Louwesweg 1,1066 EA Amsterdam, The Netherlands Adv Dent Res 7(2):158-162, August, 1993 Abstract For many years, radiographs have been a valuable aid in the diagnosis of periodontal disease and the evaluation of treatment effects. Computer-based image acquisition and processing techniques will now further increase the importance of radiography in periodontal diagnosis. Temporal changes of lesions can be made easily visible by means of subtraction radiography based on digital images. This process requires a pair of images with identical gray-level distributions and projection geometry. The gray-level distribution and perspective projection of images can be corrected by means of digital image processing. A pair of identical images can thus be obtained without mechanical alignment of patient, film, and x-ray source. Algorithms have been developed for automatical determination of the borders of lesions and can subsequently produce quantitative information ranging from simple distance measurements to advanced multidimensional quantitation of image parameters. Accurate volume measurements can be carried out by the utilization of calibration wedges in the image. Image reconstruction procedures, such as tomosynthesis, provide information about the third dimension, which is normally lost in conventional radiographic projections. The buccal and lingual sites of the alveolar crest can be inspected separately. The progress of computer-aided procedures as discussed in this paper appears to have great potential for the improvement of the radiographic diagnosis of periodontal lesions. Especially, the benefits of reproducibility and quantitative evaluation of treatment effects will greatly improve the role of radiography in periodontics. Presented at the 12th International Conference on Oral Biology (ICOB), "Modern Concepts in the Diagnosis of Oral Disease", held at Heriot-Watt University, Edinburgh, Scotland, July 6-7,1992, sponsored by the International Association for Dental Research and supported by Unilever Dental Research We acknowledge the partial funding of this study by NATO Collaborative Research Grant No. CRG 910229. The diagnosis of periodontal disease is greatly facilitated by the use of radiographs. Radiographs are helpful in the assessment of treatment effects and the prognosis of disease progress. Radiographs are considered a valuable adjunct to the clinical examination, because essential information is provided about the bony tissues covered by the gingiva that cannot be diagnosed by clinical inspection alone. Radiographic image formation is based on the principle of projecting a three-dimensional object onto a two-dimensional image plane, and therefore this technique also has limitations. The radiographic image, in principle, lacks information about the third dimension, at least in a way that is easily perceptible by the human visual system. The radiographic image gives spatial information about the object in the x and y directions only, i.e., in the plane perpendicular to the direction of the x- ray beam. Consequently, a single radiograph is inappropriate for obtaining information about this third dimension. More projections are needed for this purpose. Due to these characteristics of radiographic image formation, the orientation of the x-ray beam toward the object is a factor that is very important for the resulting x-ray view. A different orientation of the projection results in a different image, which in its turn may affect the interpretation and diagnosis based on that radiograph. It is for that reason that standardization and reproducibility are essential requirements for the reliability of the diagnosis. It is obvious that radiographs are indispensable for the diagnosis of periodontal disease, but on the other hand there are severe limitations in their abilities to provide information about the spatial relationship of anatomical and pathological entities in the direction parallel to the x-ray beam. The evolution of digital radiography will reduce the drawbacks of radiographic image formation as described in the previous paragraphs. Electronic intra-oral image receptors are now commercially available for dental applications, and the use of all-digital radiographic procedures is possible on a routine base. This paper describes some procedures that are now available, and explores some future trends in computer-based radiodiagnosis of periodontal disease. DIAGNOSTIC REQUIREMENTS In order to be useful for the purpose of diagnosis, radiographs have to satisfy the requirements of standardization and reproducibility. The ideal radiographic procedure should also facilitate the collection of quantitative data with regard to the condition of (potential) lesion areas and should provide sufficient information about the shape and the extension of the lesion in three dimensions. These requirements are even more important in the radiographic diagnosis of periodontal disease. The progress of disease symptoms is relatively slow, so the 158

VOL. 7(2) MODERN RAIOGRAPHIC METHODS IN PERIODONTOLOGY 159 Fig. 1 Example of subtraction radiography, (a) is the radiograph taken before treatment and (b) that taken after treatment. The difference between the two images is not easily visible to the naked eye. Bone ingrowth is easily visible in (c) (arrow). The region of bone gain is isolated in (d) to permit area and volume measurements. difference between subsequent radiographic views can be extremely small. The shape of a periodontal lesion in a radiograph is dependent, to a large degree, on the orientation of the radiographic projection. Comparison of radiographs taken with a time interval is possible only when the projection geometry is identical. Standardization of projection geometry is required to produce radiographic images that can be compared with 'normals'. The clinician is not able to make a reliable diagnosis unless he can compare the radiograph of a particular patient with the normal and abnormal reference images he has in his mind. Therefore, the use of film-positioning devices and the paralleling technique is strongly recommended for high-quality images to be obtained. It must be emphasized that film developing is an important factor in standardization. A careless developing procedure can interfere with the detection of small or ill-defined lesions. Reproducibility is required for reliable comparisons of radiographs taken at sequential time intervals. In most cases, this is done by connecting the x-ray machine, the patient, and the film mechanically and sometimes by using film holders provided with individual bite blocks in order to establish the exact position of the film for each examination (Duckworth et ai, 1983). It is clear that even subtle changes in the occlusal surfaces of teeth when, for instance, new restorations have been made will reduce the quality of this procedure. Jeffcoat et al. (1987) used cephalostats to overcome this problem. They report an angular disparity between repositionings of the patients within the cephalostat of as low as 0.33 ±0.1. The human visual system is very skillful in recognizing global structures, but not in detecting small details. It is impossible for the clinician to make a very accurate estimation of the absolute amount of bone loss. Other means are needed Fig. 2 Computer-aided recognition of angular periodontal bone resorption. Two lesions are visible in the interdental bone. Lesion size is corrected for the width of the periodontal ligament space, based on measurements of the average width of the periodontal ligament space in the apical part of the region of interest. The amount of bone resorption in the bucco-lingual direction is indicated by the gray value of the pixels within the lesion area (black, less bone loss; white, more bone loss). Note that this method is based on the assumption that density differences are indicative of changes in lesion volume. for the quantification of bone loss or the assessment of treatment effects when successive radiographs are compared (Hildebolt et ai, 1991). The extraction of quantitative information from the radiographs will facilitate a reliable comparison. In addition to this, methods are needed that will provide information for the clinician about the missing third dimension (parallel to the x-ray beam) as well. These methods will give either sectional views of the object or real three-dimensional representations. SUBTRACTION RADIOGRAPHY Subtraction radiography can be used to visualize small differences of bone density and bone volume over time (Grondahl and Grondahl, 1983; Okano et al, 1990). Fig. 1 demonstrates this technique. The development of subtraction radiography has been greatly facilitated by the availability of digital imaging techniques (Grondahl et al., 1983). For subtraction radiography, the requirements of standardization and reproducibility are even stronger than in conventional visual interpretation of radiographic images. The projection geometries of the two radiographs that form the pair of images used for the subtraction process in the computer need to be identical. Otherwise, the results will display the difference in registration, and this will hide the changes due to the disease process. Digital subtraction

160 VAN DER STELT radiography requires reproducibility of the projection geometry within an angle of 2 or 3 degrees (Grondahl et al., 1984). Mechanical devices have been used to maintain correct repositioning of films over time, but these methods are rather cumbersome and not useful for routine applications. Methods have been developed aiming at the reconstruction of images that have an arbitrary projection geometry into the projection geometry of a reference image (Ruttimann et al., 1986; Van der Stelt era/., 1989; Dunn and Van der Stelt, 1992). The result of the reconstruction is a pair of images with the identical projection geometries, as required for subtraction radiography. These methods eliminate the need for mechanical connection of x-ray source, patient, and film (Yen et al., 1990). Dunn and Van der Stelt (1992) report that reliable measurements can be made in dental radiographs with up to 16 mm translation errors and angulation errors of up to 32. They consider this a promising approach for creating useful image pairs for subtraction radiography. Even when the selection of exposure time and the development of the film are performed very carefully, small fluctuations of the optical densities of sets of films are sometimes unavoidable over a longer period of time. Small differences of the gray-level distribution in a pair of images can be corrected by digital image processing. Ruttimann and Webber (1986) and Ohki et al. (1988) describe software-based methods to correct the gray-level distribution of one image according to the distribution of another image. This so-called "digital gamma correction" is a convenient method for correcting the problem of different exposure conditions in apairof subtraction images. Application of this procedure, in many cases, is a prerequisite for doing reliable measurements on subtraction images. QUANTITATIVE AND COMPUTER-AIDED INTERPRETATION Changes of bone density can be measured by means of computeraided procedures. By use of a calibration wedge with known radiation attenuation properties, the measurements of density differences can be converted into estimations of volume :' \, 1 n A Am DENT RES AUGUST 1993 changes. Aluminum or hydroxyapatite is often used because of its similarity to bone in terms of radiation attenuation characteristics. In vitro studies have shown agreement within 10% between digital volume calculations and determination of the lesion size by weight (Ruttimann et al., 1985). Bragger et al. (1988) report a sensitivity of 82%, a specificity of 88%, and a diagnostic accuracy of 87% for a system measuring surgicallyinduced bone loss in humans. The results of this technique were significantly better than those obtained through visual examinations of the radiographs by a group of experienced periodontists. When the features of anatomical details and abnormalities as they are depicted on radiographic images can be expressed mathematically, then computer algorithms can be developed to interpret radiographs automatically (Ruttimann et al., 1985; Van der Stelt et al., 1985; Benn, 1991). A computerized procedure has been reported to describe the shapes and sizes of angular periodontal lesions (Fig. 2). When applied to series of radiographs taken over a certain period of time, this method can be used to follow the effects of clinical intervention, e.g., in patients treated for juvenile periodontitis. These types of computer-aided procedures can be a great adjuvant to clinical decision-making. They add objective information to the clinical assessment of the patient (Van der Stelt and Geraets, 1991). THREE-DIMENSIONAL IMAGING The local contrast in a radiographic image is the result of spatial attenuation differences of the x-rays passing through the object. Differences are caused by different materials (soft tissue, bone, etc.) and by different path lengths (e.g., thick bone layers vs. thin bone layers). The effect of both causes is the same; in many cases, it is not possible to distinguish, merely by the radiographic density, between different materials ordifferent dimensions of structures in the object. When thin layers of an object are visualized, this problem is much less important, because the effective path length through the object is reduced, and local contrast is more likely the result of different materials, corresponding to different anatomical structures. Tomosynthesis is a software-based 3 reconstruction method to produce thin layers or slices of dental structures (Groenhuis et al., ^ 1983; Ruttimann et al., 1983). The method is not yet useful under clinical circumstances, because the basis projections need to be taken according to a very strict imaging geometry (Groenhuis et al., 1984). It is beyond doubt, however, that new hardware and software solutions will be available in the near future to establish more conveniently the registration of the basis projections and make this method feasible for clinical use (Van der Stelt et al., 1989) (Fig. 3). Fig. 3 Example of two tomosynthetic reconstructions. The left reconstruction shows a cross-section of the lingual cortical plate; note the alveolar bone level (arrows) which is easily recognizable without disturbing the superimposition of other anatomical structures. The right image shows a central section with the periodontal ligament space clearly visible. THE FUTURE The progress of computer technology is very fast. It seems therefore precarious to predict the role of digital image processing in dental radiology in the distant future. Some trends,

VOL. 7(2) MODERN RAIOGRAPHIC METHODS IN PERIODONTOLOGY 161 however, may be indicative of the way dental radiography will be practiced in the near future. These trends are: automated interpretation of radiographic images to improve the objectivity of the diagnosis; development of quantitative procedures to support the diagnosis of lesion progress over time; advanced procedures for the reconstruction of radiographic images to obtain standardized and three-dimensional views. The infrastructure for high-speed transmission of large data sets, like images, will be available in many geographic areas, thus enabling dentists to exchange images, to consult experts immediately, and to use knowledge bases as an aid to diagnostic problem-solving (Stheeman et al., 1992). It is evident that the diagnosis of periodontal disease will benefit from these developments. The difficulties of visualizing the buccal and lingual plates of the alveolar process without superimposition of tooth structures will be overcome. The shapes and sizes of angular bone lesions can be judged from their three-dimensional reconstruction. Subtraction radiography for longitudinal evaluation of treatment effects can be performed on a routine basis, without the need of meticulous fixation of the patient, the x-ray source, and the image plane during the exposure. However, notwithstanding the development of all these new and promising technologies, new procedures must demonstrate their effectiveness and reasonable cost/benefit characteristics in extensive clinical trials (Goin and Hermann, 1991). Deas et al. (1991) compared probing attachment loss with digital radiographic density measurements. They found more bone loss with the digital method; in their conclusions, however, they point out that the digital method may produce more positives due to the increased sensitivity as compared with conventional (non-digital) measuring methods that have as-yet-undetermined clinical significance. Specificity, sensitivity, and other parameters of diagnostic performance must be determined and carefully evaluated before new technologies can be fully introduced in radiodiagnosis and eventually replace existing diagnostic tests. REFERENCES Benn DK (1991). Automatic analysis of radiographic images: I. Theoretical considerations. Dentomaxillofac Radiol 19:187-192. Bragger U, Pasquali L, Kornman KS (1988). Remodeling of interdental alveolar bone after periodontal flap procedures assessed by means of computer-assisted densitometric image analysis (CADIA). / Clin Periodontol 15:558-564. Deas DA, Pasquali LE, Yuan CH, Kornman KS (1991). The relationship between probing attachment loss and computerized radiographic analysis in monitoring progression of periodontitis. / Periodontol 62:135-141. Duckworth JE, Judy PF, Goodson JM, Socransky SS (1983). A method for geometric and densitometric standardization of intraoral radiographs. / Periodontol 54:435-440. Dunn SM, Van der Stelt PF (1992). Recognizing invariant geometric structure in dental radiographs. Dentomaxillofac Radiol 21:142-147. Goin JE, Hermann GA (1991). The clinical efficacy of diagnostic imaging evaluation studies; problems, paradigms, and prescriptions. Invest Radiol 26:507-511. Groenhuis RAJ, Webber RL, Ruttimann UE (1983). Computerized tomosynthesis of dental tissues. Oral Surg Oral Med Oral Pathol 56:206-214. Groenhuis RAJ, Ruttimann UE, Webber RL (1984). A prototype digital tomographic x-ray system for dental applications. Proceedings, IEEE International Symposium on Medical Images and Icons. New York: IEEE, 218-221. Grondahl H-G, Grondahl K (1983). Subtraction radiography for the diagnosis of periodontal bone lesions. Oral Surg Oral Med Oral Pathol 55:208-213. Grondahl H-G, Grondahl K, Webber RL (1983). A digital subtraction technique for dental radiography. Oral Surg Oral Med Oral Pathol 55:96-102. Grondahl K, Grondahl H-G, Webber RL (1984). Influence of variations in projection geometry on the detectability of periodontal bone lesions; a comparison between subtraction radiography and conventional radiographic technique. / Clin Periodont 11:411-420. Hildebolt F, Vannier MW, Shrout MK, Pilgrim TK (1991). ROC-analy sis of observer-response subjective rating data application to periodontal radiograph assessment. Am J Phys Anthropol 84:351-361. Jeffcoat MK, Reddy MS, Webber RL, Williams RC, Ruttimann UE (1987). Extraoral control of geometry for digital subtraction radiography. / Periodont Res 22:396-402. Ohki M, Okano T, Yamada N (1988). A contrast-correction method for digital subtraction radiography. / Periodont Res 23:277-280. Okano T, Mera T, Ohki M, Ishikawa I, Yamada N (1990). Digital subtraction of radiograph in evaluating alveolar bone changes after initial periodontal therapy. Oral Surg Oral Med Oral Pathol 69:258-262. Ruttimann UE, Webber RL (1986). A robust digital method for film contrast correction in subtraction radiography. / Periodont Res 21:486-495. Ruttimann UE, Groenhuis RAJ, Webber RL (1983). Computer tomosynthesis: A versatile three-dimensional imaging technique. Proceedings, 7th Conference of the Society of Computer Application in Medical Care (SCAMC), 783-786. Ruttimann UE, Webber RL, Groenhuis AJ, Troullos E, Rethman M (1985). Automated estimation of lesion size. Proc SPIE, Vol. 135. Bellingham (WA): SPIE, 325-330. Ruttimann UE, Van der Stelt PF, Webber RL (1986). Use of image similarity for the selection of projections for subtraction radiography. SPIE Med XIV/PACS IV, Vol. 626. Bellingham (WA): SPIE, 301-307. Stheeman SE, Van der Stelt PF, Mileman PA (1992). Expert systems in dentistry; past performance future prospects. / Dent 20:68-73. Van der Stelt PF, Geraets WGM (1991). Computer-aided interpretation and quantification of angular periodontal bone defects on dental radiographs. IEEE Trans Biomed Eng 18:334-338.

162 VAN DER STELT ADV DENT RES AUGUST 1993 Van dersteltpf, Van der Linden LWJ, Geraets WGM, Alons Van der Stelt PF, Ruttimann UE, Webber RL (1989). CL (1985). Digitized image processing and pattern Determination of projection for subtraction radiography recognition in dental radiographs with emphasis on the based on image similarity measurements. Dentomaxillofac interdental bone. / Clin Periodontol 12:815-821. Radiol 18:113-117. Van der Stelt PF, Webber RL, Ruttimann UE, Groenhuis RAJ Yen L, Dunn SM, Van der Stelt PF (1990). Finding invariant (1986). A procedure for reconstruction and enhancementof anatomical relationship in dental radiographs. Proc. IEEE/ tomosynthetic images. Dentomaxillofac Radiol 15:11-18. EMBS, Vol. 12. New York: IEEE, 2076-2077.