Room: AAPM ePoster Library
Purpose: Generally, the three-dimensional (3D) medical image is expressed by voxels with a canonical coordinate system. Although the voxel representation is easy to understand for us, but it may be redundant because a human body has a spatially high symmetry with respect to the body axis. In this study, we propose a novel reconstruction approach with spherical harmonics in a polar coordinate system. The purpose of this study is to show the feasibility and to evaluate the efficiency of this reconstruction approach.
Methods: A 3D Shepp-Logan (SL) phantom and a 3D digital head phantom were prepared and their reprojection data with a pixel size of 360 x 360 in 10-degree and 6-degree interval of gantry angle (that is, 36 and 60 reporojections) were produced by in-house ray-tracing algorithm. Using the reprojection data, the image reconstruction was then performed with an assumption that a human body can be expressed by spherical harmonics. For comparison, the Feldkamp-Davis-Kress (FDK) reconstruction and model-based iterative reconstruction (IR) method were also applied for their reprojection data. The accuracy of the reconstructed image was evaluated by a root mean square error (RMSE) with the corresponding original 3D image.
Results: The RMSE of the reconstructed 3D SL phantom with 36 projections in our proposed method was 44.9, whereas those in FDK and IR were 122.9 and 71.1, respectively. The RMSE in the head phantom was 115.9, which were also smaller than that of FDK and IR (208.6 and 132.5, respectively). The result with 60 projections had a same tendency as that with 36 projections. These results imply that the 3D reconstruction using the spherical harmonics are feasible and very efficient in comparison with the conventional reconstruction methods in under-sampling situation.
Conclusion: Cone-beam CT image reconstruction with spherical harmonics can generate high-quality images and make a human-body expression efficient.