| Sign In to gain access to subscriptions and/or personal tools. |
Synthesizing Dental Radiographs for Human Identification
1 School of Information Technology and Electrical Engineering, The University of Queensland, Brisbane QLD 4072, Australia; and Correspondence: * corresponding author, mehnert{at}itee.uq.edu.au
The task of identifying human remains based on dental comparisons of post mortem (PM) and ante mortem (AM) radiographs is labor-intensive, subjective, and has several drawbacks, including: inherently poor image quality, difficulty matching the viewing angles in PM radiographs to those taken AM, and the fact that the state of the dental remains may entirely preclude the possibility of obtaining certain types of radiographs PM. The aim of the present study was to investigate the feasibility of using radiograph-like images reconstructed from PM x-ray computed tomography (CT) data to overcome the shortcomings of conventional radiographic comparison. Algorithms for computer synthesis of panoramic, periapical, and bitewing images are presented. The algorithms were evaluated with data from clinical examinations of two persons. The results demonstrate the efficacy of the CT-based approach and that, in comparison with conventional radiographs, the synthesized images exhibit minimal geometric distortion, reduced blurring, and reduced superimposition of oral structures.
Key Words: computed tomography • radiograph • Radon transform • dental comparison • forensic identification
Dental comparison provides one of the best avenues for the forensic identification of human remains, because teeth, unlike other body parts, are highly resistant to decomposition and are not easily destroyed. Post mortem (PM) dental information is compared with ante mortem (AM) information in an attempt to establish the identity of the remains. Dental information typically includes x-ray-based radiographs—including periapical film and bitewing radiographs, and panoramic radiographs or orthopantomographs (OPGs)—and written notes and charts from patient records. Dental comparison is based on characteristics of the teeth, jaws, and bucco-lingual structures, as well as on the characteristics of any restorations or dental work (Adams, 2003). The classic approach to dental identification involves manually summarizing dental records and radiographs. A PM summary can then be compared with AM summaries to establish a short list of possible identities. Manual comparisons of the candidate AM radiographs with the PM radiographs are necessary before a positive identification can be made. The classic approach to dental comparison has several shortcomings. First, it is very labor-intensive (Kogon et al., 1974). Second, it is error-prone, because mistakes can be made when information is collected and summaries are compiled (Kogon et al., 1974). Third, it is very subjective (Kogon et al., 1974). Finally, and perhaps most significantly, the short list of possible matches is based on summary information rather than on features, e.g., tooth and root morphology, derived directly from the PM and AM radiographs. These shortcomings have motivated research into, and development of, computer-assisted methods to improve the accuracy and efficiency of dental identification. These fall into two categories: (i) methods that aim to improve the efficiency and accuracy of the classic approach; and (ii) methods that aim to extend the classic approach. Software packages such as CAPMI (Lorton et al., 1988), WinID (McGivney, 2006), IDIS (Chomdej et al., 2006), and OdontoSearch (Adams, 2003) belong to the first category. These packages essentially permit the creation of AM/PM computer databases using standardized summary forms, and text-based searching of records. With regard to the second category, several approaches to content-based matching of digitized radiographs have been proposed (Jain et al., 2003; Jain and Chen, 2004; Chen and Jain, 2004; Said et al., 2004; Zhou and Abdel-Mottaleb, 2004, 2005; Mahoor and Abdel-Mottaleb, 2005; Nomir and Abdel-Mottaleb, 2005). These approaches utilize image processing and statistical pattern recognition techniques to perform automated or semi-automated matching of digitized radiographs based on tooth shape. The majority of existing approaches to computer-assisted dental comparison have focused on the use of conventional x-ray radiographs. The reason for this is that conventional radiographs are the most abundant source of AM image data. However, this approach has several drawbacks, including: (i) reduced image quality in both AM and PM images, as a result of improper film processing and geometric distortion; (ii) difficulty discerning tooth structure because of overlapping adjacent teeth and/or other oral structures; (iii) difficulty matching views in corresponding PM and AM radiographs, because of differing projection geometries at the time of acquisition; and (iv) the fact that the quality of the remains may preclude the possibility of acquiring certain types of radiographs PM. An alternative approach, which overcomes these drawbacks, is to use PM x-ray computed tomography (CT) for matching against AM radiographs (Jackowski et al., 2006; Thali et al., 2006). CT is an imaging modality that reveals the three-dimensional (3D) relationships between teeth and their supporting tissues in a series of two-dimensional (2D) digital images without geometric distortion. Unfortunately, CT is not an abundant source of AM dental image data. This is because of the higher cost and the higher dose of radiation to which a person is exposed compared with conventional radiography. Nevertheless, CT data can be easily acquired PM and, given that CT is a true 3D modality, used to generate 2D images that mimic conventional radiographs. Moreover, these images can be generated with arbitrary projection geometry, reduced blurring, and reduced superimposition of other oral structures. The aim of the present study was to investigate the feasibility of this approach. Algorithms for computer synthesis of panoramic, periapical, and bitewing radiograph-like images from CT data are presented. In addition, an evaluation of the algorithms, using real clinical data, is presented.
Overview of Conventional Dental Radiography Periapical and bitewing radiographs are the most commonly acquired radiographs in clinical practice. These are referred to as ’intra-oral’ radiographs because the film is placed in the person’s mouth. They are used to assess dental caries. A periapical radiograph is used to examine individual teeth and the tissue surrounding the root apices (Whaites, 1996). Each film typically presents from 1 to 4 teeth. Bitewing radiographs are named after the original technique, which required the person to bite on a small wing attached to the intra-oral film. This technique is used to examine the crowns of both the premolar and molar teeth on one side of the jaw (Whaites, 1996). Three or four upper and lower tooth crowns are typically displayed in each film. Both periapical and bitewing radiographs aim to record a select range of teeth with the least magnification error. The ideal imaging geometry requires the film and the teeth to be parallel to one another, and for parallel x-ray beams to be transmitted perpendicular to the teeth and film (Whaites, 2002). The orientation of the person, film, and x-ray tube head are fixed during the exposure of the film. Consequently, given the limits in film positioning because of oral anatomy, in some cases the film may not be in the ideal position, leading to geometric distortion in the resulting image. A panoramic radiograph, or OPG, is an extra-oral radiograph used to supplement bitewing and periapical radiographs. In panoramic radiography, both the x-ray tube head and the film are simultaneously moved around the person’s head during the exposure. The process of image formation is known as tomography. The x-ray source emits a narrow vertical beam, and as the system rotates, a 2D image is exposed on film. The method permits a single curved layer through the jaws to be imaged, while at the same time blurring out shadows of oral structures superimposed from other layers. This curved layer is known as the ‘focal trough’ or ‘image plane’. Manufacturers design the size and shape of the focal trough to accommodate the average jaw.
Overview of X-ray Computed Tomography
Image Projection and the Radon Transform
Algorithms for Reconstructing Radiograph-like Images from CT Data
Our proposed algorithm for reconstructing periapical- and bitewing-like images from CT data (Table 1  ) requires that the user manually define a 3D region of interest (rectangular prism) around the tooth or teeth that are to appear in the projection.
Our proposed algorithm for reconstructing OPG-like images from CT data (Table 2
Evaluation Using Routine Clinical Data Retrospective anonymized image data for two persons were obtained from a local dental imaging practice. The data were chosen to include a person for whom a spiral CT dataset and an OPG film had been acquired (Person A), and a person with no metallic restorations (Person B). Approval from the Human Research Ethics Committee of the University of Queensland was obtained for this study. The CT data for Person A consisted of 109 axial slices through the mandible (voxel size of 0.24 mm x 0.24 mm x 0.50 mm) and 115 axial slices through the maxilla (voxel size of 0.18 mm x 0.18 mm x 0.50 mm). The CT data for Person B consisted of 113 axial slices through the mandible (voxel size of 0.22 mm x 0.22 mm x 0.50 mm). The reconstruction algorithms were implemented in MATLAB 7.1 (The MathWorks Inc., Natick, MA, USA, 2005). The algorithm for reconstructing periapical- and bitewing-like images was applied to the CT data for Person A with a variety of user-defined regions of interest. The algorithm for reconstructing an OPG-like image was applied to the combined mandible and maxilla CT data for Person A, and to the mandible-only CT data for Person B.
The reconstructed periapical- and bitewing-like images (Fig. 2a  ) revealed tooth composition, dental caries, dental restorations, and, in the case of the periapical images, also the root shape and supporting bone in the anterior teeth. Negatively, the reconstructed images contained intensity artifacts as the result of metal artifacts in the original CT data, originating from metallic dental restorations. The presence of these artifacts can make interpretation difficult (e.g., Fig. 2a, bottom-left).
In the case of Person A, a visual comparison between the reconstructed OPG (Fig. 2b ) and the real OPG (Fig. 2c) revealed that the former exhibited reduced geometric distortion, reduced blurring, and reduced overlapping of dental structures. Of particular note was that in the reconstructed OPG: (i) there was less horizontal magnification in the premolar area; (ii) there was minimal overlapping of the posterior teeth; (iii) the root curvature of the permanent lower molars was preserved; (iv) the dental restorations in both the crown and root portions of restored teeth could be seen clearly; and (v) there was no superimposition of other oral tissues. Negatively, the reconstructed OPG contained intensity artifacts stemming from metal artifacts manifest in the CT data. Similarly, in the case of Person B, a visual comparison between the reconstructed OPG (Fig. 2d) and the CT volume rendering (Fig. 2e) revealed that the former exhibited little geometric distortion, blurring, and overlapping of dental structures. The reconstruction, in this case, was free from intensity artifacts, because the person had no metallic restorations.
The results demonstrated that it is possible to reconstruct 2D images from spiral CT data that mimic conventional periapical-, bitewing-, and panoramic radiographs. In addition, they demonstrated the efficacy of the proposed computer algorithms for reconstructing these images. Collectively, the results demonstrated that it is indeed feasible to use these images in lieu of conventional radiographs for matching against real AM radiographs, to establish the identity of human remains. This approach to dental comparison opens up the possibility for automated, objective comparison by computer. The approach overcomes several limitations of conventional radiographic comparison, including: inherently poor image quality, difficulty matching the viewing angles in PM radiographs to those AM, and the fact that the state of the dental remains may entirely preclude the possibility of obtaining certain types of radiographs PM. Metallic dental restorations lead to streak-like intensity artifacts in the CT image data. These occur because x-rays that pass through metal objects are much more attenuated than those that pass through soft tissues and bone. These artifacts, in turn, lead to intensity artifacts in the reconstructed radiograph-like images produced by our algorithms. They occur, in particular, on and around teeth with restorations. They affect not only the visual quality of the reconstructions, but also their interpretability. For this reason, we intend to incorporate metal artifact reduction as a pre-processing step in future versions of our reconstruction algorithms. This research represents the first step in a larger initiative to develop a computerized image analysis system to assist the forensic dentist with the task of identifying human remains. In particular, the system will permit radiograph-like images to be reconstructed from PM CT data dynamically ("on the fly"), and will enable these to be matched against AM digitized conventional radiographs. Matching will be performed on the basis of quantitative features extracted from these images that characterize both the morphology of the teeth and jaws and their spatial interrelationships.
Ms. Tohnak acknowledges financial support from Naresuan University and the Commission on Higher Education, Ministry of Education, Thailand. Received for publication October 17, 2006. Revision received June 28, 2007. Accepted for publication July 9, 2007.
Journal of Dental Research, Vol. 86, No. 11,
1057-1062 (2007)
|
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
|
|
 |
|
Sign In to gain access to subscriptions and/or personal tools.
TOP
ABSTRACT
INTRODUCTION
