Development of fast patient position verification software using 2D-3D image registration and its clinical experience

摘要

Treatment workflow for carbon-ion beam treatment is similar to that for proton beam and external beam photon therapy. The design of our new treatment facility was focused on providing a treatment workflow that could be adapted for use in any treatment situation. The main workflow consists of immobilization, CT acquisition, treatment planning, simulation, quality assurance, and treatment beam irradiation. The ITG includes two functions, 2D/2D manual registration (2D/2D-ITG) and 2D/3D auto-registration (2D/3D-ITG). These two functions work interactively. Here, we focus on 2D/3D-ITG. DRR calculation is the summation of CT voxel values along the X-ray projection ray. Although several DRR calculation algorithms have been reported [id="xref-ref-8-1" class="xref-bibr" href="#ref-8">8, id="xref-ref-9-1" class="xref-bibr" href="#ref-9">9], our implementation [id="xref-ref-10-1" class="xref-bibr" href="#ref-10">10] was designed to improve calculation speed based on the extension of a Siddon ray tracing algorithm [id="xref-ref-9-2" class="xref-bibr" href="#ref-9">9] to volume data using trilinear interpolation to reduce interpolation errors [id="xref-ref-7-2" class="xref-bibr" href="#ref-7">7]. To provide similar image quality to FPD, we convert these CT voxel data to the X-ray attenuation for photon energy using image processing of the CT-number weighting in each CT voxel data before integration of the pixel value along the ray. Another fast calculation technique segments the CT data within the patient surface and integrates CT voxel values within this region only. Moreover, we set calculation ROIs on orthogonal FPD images, which are user-selected registration regions, on the basis that patient anatomical position may not be exactly the same in respective treatment fractions [id="xref-ref-10-2" class="xref-bibr" href="#ref-10">10]. If the user forgets to set the calculation ROI, the ITG automatically sets it by expanding the target ROI regions. DRR calculation is done within the above regions only. These calculation regions are recorded in an RT image, and automatically displayed during the treatment course. These two techniques each provide a significant reduction in computation cost. Parameters described here are DRR projection parameters (six degrees of freedom (6DOF)), which iteratively estimate the unknown pose of the X-ray system relative to the CT volume. These parameters are defined using normalized mutual information (NMI) [id="xref-ref-11-1" class="xref-bibr" href="#ref-11">11], gradient differences (GDs) [id="xref-ref-5-2" class="xref-bibr" href="#ref-5">5], and zero-mean normalized cross-correlation (ZNCC) as similarity measures (see Appendix). We used a single or combination of these metrics to calculate the final score by applying weighting values to the respective scores for each anatomical site. 2D/3D-ITG calculates the four respective parameters from each image direction, namely the registration errors for X, Y, φ and θ and those for Y, Z and ψ, as derived from the vertical and horizontal images, respectively. This is because positional errors in translation and rotation on the plane are more easily recognized than those out of the plane (Fig. id="xref-fig-1-4" class="xref-fig" href="#F1">1). For example, positional errors in Y, Z and ψ are easy to recognize in horizontal images. Therefore, weaker directions (e.g. X and θ in horizontal images) in respective images are supported by each other. Since vertical and horizontal images share a Y-axis, registration errors in Y and φ derived from respective images are averaged. This averaging allows the calculation of 6DOF registration errors. However, when two FPDs are installed in any configuration apart from 90°, 2D/3D-ITG calculates the six parameters from each image direction. Orthogonal FPD images are acquired in two sets using an X-ray imaging system (Canon CXDI-55C, Tokyo, Japan) with an imaging area size of 35 cm × 43 cm and a pixel pitch of 0.16 mm. A CsI scintillator receptor is used for static image acquisition. All FPDs are installed within the port cover. The vertical X-ray tube is set under the floor (Fig. id="xref-fig-2-2" class="xref-fig" href="#F2">2), and the horizontal X-ray tube is set at the opposite side of the horizontal FPD and moved down when it is used. The distance from the room isocenter (ISO) and source–image receptor distance (SID) are 155 cm and 213 cm, respectively. We evaluated the 2D/3D-ITG function in terms of computation time and registration accuracy. Registration errors were defined using the similarity metrics GD, NMI, ZNCC and their combination. These metrics have various characteristics, as described in the Appendix and can be strongly affected on the image quality (bone emphasizes sites such as pelvic and head regions and s