|
|
Advanced signal processing put at the service of caries diagnosis This publication appears in: DSP Valley Newsletter Authors: B. Truyen Volume: 3 Issue: 1 Pages: 7-7 Publication Year: 2002
Abstract: Dental caries, the most common among all diseases, generally is not regarded as one of the major health issues of this century, despite its enormous economic impact on the public health care system. Being a bacterial induced disease, caries starts as a localized demineralization of the outer enamel layer, and progressively expands towards the inner dentine tissue, eventually leading to extensive cavitation. Given the treatment ramifications of advanced dentinal caries, early diagnosis is of prime importance. The inefficiency of current diagnostic tools, such as visual assessment possibly supplemented with bite-wing radiographs, with sensitivities as low as 50 %, has launched intensive research for more performant methods.One such method, which is much more sensitive to subtle caries induced changes in radiolucency between radiographs acquired separated in time, is digital subtraction radiography (DSR). Yet, DSR has only seen sporadic clinical acceptance for the diagnosis of slowly progressing caries. This reluctance certainly is related with the difficulties in achieving reproducible geometric conditions between radiographic exposures. Instead of relying on mechanical stabilization devices to replicate the relative position of source, tooth, and intra-oral radiographic film, mathematical methods are being developed, which allow the necessary geometric corrections to be performed retrospectively. Compared to other uses of DSR, its application to caries diagnosis poses some particularly challenging problems, following the extremely high accuracy requirements. Indeed, the location of caries lesions on the outer surface of the enamel layer, combined with the very high radiographic contrast between the enamel layer and the air gap between teeth, makes that even the slightest geometric misalignment may cause a complete obfuscation of the radiolucency.Within VUB-ETRO, research has been initiated on the application of the information theoretic concept of mutual information (MI) for retrospective geometric correction (1). Unlike all previous approaches that are based on the (automatic) identification of a (possibly large) number of corresponding feature points, this automated method recovers the optimal geometric correction from the gray value distributions. Besides the convenience of dispensing with the tedious and error prone manual identification of feature points, the method shows a better accuracy than existing manual and automatic methods.Also, improvements upon existing reference point methods are further studied within VUB-ETRO, with a particular emphasis on consistent parameter estimation concepts. All currently applied ordinary least-squares based algorithms make the tacit assumption that reference points in one of the two images are identified without error margins. Instead, Total-Least Squares (TLS) allows to account correctly for the symmetrical situation underlying the image registration problem, in which reference points in both images are subject to positional errors.Based on the significant differences between the frequency dependent electrical properties of healthy and carious dental tissue, several techniques for the assessment of incipient caries have been proposed in past years. These methods, which basically are 2-point impedance measurement techniques, unfortunately have been shown to suffer from some conceptual limitations with regard to their accuracy. Research conducted within VUB-ETRO seeks to combine the sensitivity of the established method of Electrical Impedance Spectroscopy (EIS), with some of the cross sectional imaging features provided by the use of tiny electrode arrays placed on the tooth surfaces. Such a tomographic imaging technique, also known as Electrical Impedance Tomograpy (EIT), previously has been investigated for various other biomedical applications, but never before has been applied to the imaging of hard tissue. The major issues in this novel application are related to the very high impedance values that are typically measured for healthy tooth tissue. The accompanying raise in measurement noise calls for adequate signal processing algorithms. Emphasis in this research is especially on the use of so-called sub-space algorithms, which are known to yield much improved robustness properties.
|
|
