Aim: Estimation of morphological parameters of the heart from nuclear cardiology images such as gated SPECT or PET is usually performed by model-based fitting of radial 1D activity profiles through the left ventricle (LV) wall after image reorientation in short-axis. In this study, we evaluate a new data rebinning scheme based on 3D prolate spheroidal coordinates. The proposed method does not require image reorientation and hence it is less prone to suffer from image degradation due to voxel interpolation; furthermore, the method is robust with respect to noise because it employs all the voxels in the reconstructed images instead of just line profiles. Materials and Methods: The frame of reference of the LV with respect to the scanner is first defined by selecting three points in the reconstructed image: the center of the LV base (P1), the apex (P2) and the center of the RV cavity (P3). Once the above three points are defined, 3D prolate spheroidal coordinates are computed for each voxel; afterwards, activity profiles are obtained by partitioning the half-space containing the heart in several solid sectors, each covering a small interval of the azimuthal and radial coordinates. Angular intervals were kept small enough to avoid inhomogeneity within each sector, but ensuring a sufficient number of points within it. The curvilinear activity profile of each sector was then fitted to the common 1D model to find optimal values of LV cavity radius (R) and wall thickness (d). Anisotropy and shift-variance of the PSF are also included in the model. The proposed method was tested on computational phantoms with various levels of additive gaussian noise and realistic anisotropic PSF and voxel size. Initial tests were also performed on gated PET images of healthy patients. Results and Conclusion: Good agreement was observed between measurements on the noisy phantom and the ground truth. Both the LV cavity volume and the myocardial thickness for each sector were within 10% of the true value. In some sector, the fit failed to converge probably because of a bad choice of initialization parameters. In such cases, bilinear interpolation was used to recover the values of R and d. The proposed method is alternative to standard cardiac analysis based on image reorientation and linear profiles, avoiding interpolation and providing robustness due to the full data utilization. Furthermore, the prolate spheroidal coordinates are optimal because they minimize the angle of incidence to the endocardial surface.
Interpolation-free 3D data rebinning in prolate spheroidal coordinates for robust estimation of morphofunctional parameters in nuclear cardiology
Tripodi M;Panetta D;Salvadori P
2013
Abstract
Aim: Estimation of morphological parameters of the heart from nuclear cardiology images such as gated SPECT or PET is usually performed by model-based fitting of radial 1D activity profiles through the left ventricle (LV) wall after image reorientation in short-axis. In this study, we evaluate a new data rebinning scheme based on 3D prolate spheroidal coordinates. The proposed method does not require image reorientation and hence it is less prone to suffer from image degradation due to voxel interpolation; furthermore, the method is robust with respect to noise because it employs all the voxels in the reconstructed images instead of just line profiles. Materials and Methods: The frame of reference of the LV with respect to the scanner is first defined by selecting three points in the reconstructed image: the center of the LV base (P1), the apex (P2) and the center of the RV cavity (P3). Once the above three points are defined, 3D prolate spheroidal coordinates are computed for each voxel; afterwards, activity profiles are obtained by partitioning the half-space containing the heart in several solid sectors, each covering a small interval of the azimuthal and radial coordinates. Angular intervals were kept small enough to avoid inhomogeneity within each sector, but ensuring a sufficient number of points within it. The curvilinear activity profile of each sector was then fitted to the common 1D model to find optimal values of LV cavity radius (R) and wall thickness (d). Anisotropy and shift-variance of the PSF are also included in the model. The proposed method was tested on computational phantoms with various levels of additive gaussian noise and realistic anisotropic PSF and voxel size. Initial tests were also performed on gated PET images of healthy patients. Results and Conclusion: Good agreement was observed between measurements on the noisy phantom and the ground truth. Both the LV cavity volume and the myocardial thickness for each sector were within 10% of the true value. In some sector, the fit failed to converge probably because of a bad choice of initialization parameters. In such cases, bilinear interpolation was used to recover the values of R and d. The proposed method is alternative to standard cardiac analysis based on image reorientation and linear profiles, avoiding interpolation and providing robustness due to the full data utilization. Furthermore, the prolate spheroidal coordinates are optimal because they minimize the angle of incidence to the endocardial surface.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
