Publication Details
Cristina Boca, Bart Truyen, Adam Wasek, Lucia Henin, Walter Serneels, Peter Bottenberg, Jan Cornelis

Journal of Dental Research

Contribution To Journal


Objectives: To evaluate the effect of angular misalignment due to object rotation on the performance of a mutual information theory based method for retrospective geometrical standardization, and to compare these results with those obtained using correspondences between manually or automatically selected pairs of discrete feature points.Methods: Bitewing radiographs of the molar region of a dry human skull were acquired using a Sidexis CCD system (Sirona, Bensheim, Germany). The effect of object rotation on the performance of each method was assessed by computing the standard deviation of the difference images (SDDI) obtained by subtracting geometrically standardized images from the 0° reference image, for angular positions ranging from 8° mesial to 8° distal. This produces geometric inconsistencies similar to those observed in a clinical setting, but violating the mathematical conditions for reversibility of the image transform. Performance of the manual method was calculated by averaging over the results of 3 different operators.Results: For all angulations evaluated, the mutual information based method was found to improve significantly upon the performance of the two other methods (paired t-test, p<0.001), with SDDI values reduced by as much as a factor 4. Differences between methods based on manual or automatic selection of discrete features were only marginally significant for distal rotations, and failed to reach statistical significance for mesial rotations (p>0.05 at most angles). Strong correlations were demonstrated between object angulation and SDDI values (Spearman rank correlation, r<sub>s</sub>>0.98) for all methods evaluated. Angular dependency, however, was found to be significantly lower for the mutual information based method than for both other methods.Conclusions: The results of this study suggest that, unlike correspondences between discrete image features, the information theoretic concept of mutual information might serve as a robust basis for retrospective geometric standardization of bitewing radiographs, also in case of irreversible image distortion.