Pd-103 Strings: A New Approach to Brachytherapy Source Design

Abstract

corresponding, which ensures one-to-one point correspondence after transformation. In this study, we chose the bladder surface in the first fraction (named F1) as a reference. The bladder surfaces in the remaining fractions (named F2-F5) were transformed onto the reference in order to obtain the deformation vector fields (DVFs) on the mesh vertices. As a comparison, the DVFs were also calculated through an inverse-consistent Demons algorithm using the same binary images. The accuracy of DVFs on bladder surface vertices were assessed with several metrics: the vertexto-vertex distance (VVD), defined as the Euclidean distance of the vertices on the deformed surface to their closest vertices on the target surface, the Hausdorff distance (HD) between the deformed surface to the target surface, the percent error of the deformed bladder volume from the expected volume (PE), and the conformity index (CI) between the deformed and the target bladder volume. Better results can be indicated by lower VVD, HD, and PE, or higher CI. Results: As an example, the overlapped surface meshes (F2 to F1) through rigid registration, Demons, and TPS-RPM are illustrated in Figure 1. While it shows that Demons performed on binary images can only match the two surface vertices to a certain degree, TPS-RPM is able to achieve more accurate alignment. The performance of TPS-RPM is further quantitatively compared with Demons using VVD, HD, PE and CI when deforming F2-F5 to F1. The mean VVD and its corresponding standard deviation for the four deformations are 4.6( 3.0)mm after Demons, and 1.9( 1.4)mm after TPS-RPM. The mean HD for these four cases is 18.2mm after Demons, and 5.4mm after TPS-RPM. The mean PE is 77.0% after Demons and decreases to 16.9% after TPSRPM. The mean CI is 0.64 after Demons, and increases to 0.91 after TPS-RPM. Both visual inspection and quantitative evaluation results indicate that TPS-RPM is more accurate in performing the deformation between two surfaces Conclusion: The bladder wall deformation is more accurate using the feature-based TPS-RPM algorithm than the intensity-based Demons algorithm, indicating that TPS-RPM has the potential for accurate bladder dose deformation and dose summation for multi-fractional cervix HDR brachytherapy