Intraoral digital subtraction radiography, and its application to caries detection: Final report project "Intraoral digital subtraction radiography, and its application to caries detection"

Authors: B. Truyen

Publication Date: Apr. 2006

Abstract:

While in most western countries the prevalence of caries has decreased in the past decades, clinical diagnosis by visual examination has become more problematic. These findings are indicative of the difficulties that exist among present populations in detecting carious lesions that may progress into dentin without a macroscopically visible enamel surface breakdown. In recent years, bitewing radiography has been considered as a significant adjunct to clinical examination. However, the effectiveness of the conventional radiographic method in detecting caries has been questioned. Gratt et al. and Douglas et al. concluded from their study that the sensitivity of bitewing radiographs in detecting incipient dental caries is only a disappointing 60% or less, while the actual depth of a carious lesion is consistently underestimated. In view of the treatment ramifications of advanced caries, sensitive diagnostic methods that allow for a minimal invasive therapy are of crucial importance. Digital subtraction radiography (DSR) has been established as a sensitive technique for the detection of minute changes in serial radiographs that are difficult, if not impossible, to detect from direct visual comparison. Provided that a consistent exposure geometry - that is the relative position of X-ray source, object, and imaging sensor, can be ensured between successive radiographs - DSR can markedly improve the detection of hard tissue changes such as those associated with periodontal lesions as well as bone regeneration therapy. Nummikosky et al. found that DSR was superior to conventional radiography in detecting mechanically prepared artificially recurrent caries lesions, mainly by improving upon the specificity of the diagnoses. Furthermore, DSR also has been demonstrated to allow the in vitro visualisation of re- and demineralisation effects of caries-like decalcifications in longitudinal caries progression studies. The improved sensitivity and reproducibility in the detection of subtle alterations in density as provided by DSR, has been attributed to its ability to cancel the invariant image structure that is projected over the actual pathological changes so that these become more conspicuous. Caries initiation sites are located in comparatively homogenous image areas containing few disturbing structures. Even so, relatively high differences in mineral content were found necessary to detect the progression of a carious lesion by visual comparison of serial radiographs. This apparent inconsistency might be explained by the radiographic appearance of the demineralized area corresponding to a carious lesion, which is not a well-defined radiolucency, but instead shows a distinctive gradient of decreasing optical density from the outer enamel surface towards the dentinoenamel junction. Nevertheless, DSR still seeks general acceptance in clinical practice. This apparent reluctance may be attributed to two aspects of the DSR procedure as studied so far. The greatest obstacle to the application of DSR as a clinical caries detection tool arises from the difficulty in securing a constant exposure geometry between successive radiographic exposures. Errors in positioning are especially noticeable due to the location of the carious lesion on the outer surface of the enamel layer and the very high radiographic contrast that exists between the enamel layer and the air gap between teeth. Consequently, even the slightest geometric misalignment between images may cause a white enamel edge in one image to become superimposed over the black air gap of the paired image. This may produce undesirable artifacts that can mimic or even obfuscate true changes in radiolucency. It is possible to achieve a consistent imaging geometry using a mechanical system of occlusal stents and cephalostats, but this becomes cumbersome to use clinically as these devices must be custom-made for each patient and the entire procedure requires utmost precision. Furthermore, stents have been found only to ensure accurate relocation for an interval of maximum one year between subsequent radiographs. This effect is due to a deterioration in the fit of the stents caused by tooth mobility. When a patient tries to position a badly fitting stent on its teeth, angulation errors can arise. Consequently, patients with advanced periodontitis would be excluded altogether from the DSR procedure. Differences in exposure geometry can to some degree be corrected by the application of a retrospective geometrical standardisation, which relies on numerical transformations to bring radiographic pairs into spatial correspondence. Moreover, it has been demonstrated that even when mechanical stabilisation aids are used, significantly more accurate results are obtained when in addition retrospective geometrical standardisation is applied. Another limitation as to the further proliferation of DSR stems directly from the nature of currently used manual methods for retrospective standardisation. These methods depend on the manual assignment of corresponding feature points in pairs of radiographs to compute the actual spatial transformation so as to spatially align them. The assignment process requires considerable precision and dexterity on the part of the operator and may introduce considerable variability. The latter aspect probably is to be related to the difficulty to locate accurately corresponding anatomical feature points in radiographs acquired with an inconsistent exposure geometry. It is highly unlikely that a method requiring such an extensive operator involvement would ever find acceptance in a clinical setting as a tool for the diagnosis of caries. The aim of this research is to advance the development of digital subtraction radiography as a clinical tool for the detection of incipient caries, through (i) the development of a new instantaneous diagnostic procedure in which the application of DSR is combined with the use of a radiopaque contrast compound (ii) the derivation of several numerical contributions to improve the robustness and consistency of automatic retrospective geometrical image standardisation. The uptake of chemical elements with a high atomic number in demineralized dental hard tissue, after the topical application of a radiographic contrasting compound, has been shown to lead to an increased radiographic density. When assessed on conventional radiographs, the increase of radiopacity may actually obscure the carious lesion and therefore is not regarded as a valid clinical sign of caries. Subtraction images made from radiographs acquired before and after the topical application of a contrast compound, however, will reveal the deposit as a distinct intensity increase. In an in vitro experiment, the uptake of stannous fluoride has been shown to disclose demineralizations not discernible by direct visual inspection of conventional radiographs. It was also found that the uptake of a radiographic contrast compound altered the appearance of the subtraction image to such a degree that demineralized areas become more conspicuous so that their extent can be determined unequivocally. The combination of DSR with the topical application of a radiographic contrasting compound may thus well provide a sensitive diagnostic procedure for the early detection of caries. A procedure like this would dispense with many of the practical problems associated with the use of DSR in the longitudinal study of caries progression. Since the time interval between successive exposures should only allow for the uptake of the radiographic contrast compound, ensuring a constant imaging geometry is much easier to accomplish. In developing a taxonomy of algorithms for the retrospective geometrical standardisation of radiographs taken with a nonconsistent imaging geometry, it is common to distinguish between feature-based algorithms, that derive a geometrical transformation by minimising the distance between homologous features, and the more recent pixel-based methods, that directly use the similarity between the intensity values of corresponding pixels. Research into both classes of algorithms will be pursued, but rather than to derive yet another variant on an existing geometrical standardisation algorithm, emphasis in the proposed study will be on the contribution of specific numerical improvements to some existing and highly regarded algorithms. With regard to the use of the pixel-based algorithms, emphasis will be on the concept of Mutual Information (MI) for which a new economic global optimisation method, is going to be introduced, that derives much of its features from the properties of radial basis functions. This optimisation method will address both the problems caused by the occurrence of interpolation artefacts in the registration functional, and the need to reduce the significant computational cost of the original MI-based method. It will also be investigated how the concept mutual information can benefit feature-based algorithms in solving the very complex problem of finding correspondences between sets of unlabeled features. Finally, the asymmetry introduced in calculating the geometrical transform by minimising the distance between pairs of discrete features using classical least-squares method is going to be investigated. It has to be evaluated to what extent the use of consistent parameter estimation methods, like TLS, can eventually improve the accuracy of geometrical standardisation.