Increase Image Resolution or Acquisition Speed. Dana C. Peters, Charles A. Mistretta, Frank R. Korosec, James Holden, Fred Kelcz, Kristin L. Wedding.
Using Projection Reconstruction with a Limited Number of Projections to Increase Image Resolution or Acquisition Speed Dana C. Peters, Charles A. Mistretta, Frank R. Korosec, James Holden, Fred Kelcz, Kristin L. Wedding and Thomas M. Grist University of Wisconsin Departments of Physics, Medical Physics and Radiology Madison, Wisconsin USA
Introduction In spin warp imaging, imaging smaller fields of view (FOV) permits a reduction in scan time for a fixed resolution, or an increase in resolution for a fixed scan time. However, objects outside the FOV will appear in the small FOV at the wrong location. Methods for reduced FOV imaging have been reported [1, 2]. Here we present projection reconstruction (PR) with a limited number of projections, which offers the advantages of small FOV imaging at the expense of introducing artifacts [3] that may be tolerable. In order to satisfy Nyquist’s sampling theorem in PR, the number of projections N p required is given by Np = Nx*π/2, where Nx is the diameter of the FOV in pixels [4]. We have investigated the use of PR in which the number of acquired projections is reduced relative to a fully sampled PR acquisition. The result of reducing the number of projections by a factor of 4 is simulated in Fig. 1A. This is a 256x256 image reconstructed using 64 projections. The objects are adequately reconstructed with spatial resolution determined by the frequency-encoding (readout) resolution. But the objects contribute artifacts (which has been named “clutter” in CT literature [5]) everywhere except within a certain region about the given object. The diameter, in pixels, of the clutter-free “small FOV” centered at any object is Nfree= 2Np/π. If Nfree is equal to the readout resolution, Nx, no clutter appears anywhere in the image (fully sampled PR). Fig. 1B shows an image in which a smaller object is adequately reconstucted using 64 projections, and the artifacts caused by undersampling are outside of the “small FOV” region having a diameter Nfree. Fig. 1A indicates that objects in one location may contribute clutter to the otherwise well reconstructed information at another location. Thus, as number of projection angles is reduced below Np = Nx*π/2 , the same spatial resolution as a full angle reconstruction is locally obtained at each point in the image at the expense of increased artifacts. These artifacts may be acceptable in selected clinical applications.
A
Figure 2: Spin warp (left) vs. PR (right) at same readout resolution and scan time. PR improves resolution. FOV= 16 cm. Figure 3 compares 3D MIP images of the circle of Willis. A 512x128 spin warp image is shown on the left. The image on the right is a 512 x 128 angle PR obtained in the same time with identical imaging parameters: 32 slices zero-filled to 64, 1.4 mm thick slices, TR/TE =33/3 ms, flip 25°, 16kHz BW, 20 cm FOV. Although the SNR is decreased due to the smaller true pixel dimension of the projection image, the vessels are considerably sharper in the horizontal direction. Artifacts in the head do not seem to be a significant problem even with a factor of four reduction in the acquired number of angles compared to fully sampled PR.
B
Figure 1. Simulation using 64 projections to reconstruct 256x256 images.
Figure 3. Comparison of 3D MIP of the circle of Willis using spin warp (left) and PR (right). Same scan time and imaging parameters.
Methods A hybrid 3D fast gradient echo sequence was created from a conventional spin warp 3D sequence to acquire k-space projections in the x-y plane and slice-encodings in the z direction. The projections were acquired using fractional echoes, at multiple angles rotated through 180°. The pulse sequence was implemented on a 1.5T General Electric Signa scanner. A similar 2D sequence was used to image a resolution phantom. The reconstruction was performed off-line using a filtered back projection algorithm. Comparison images were collected using the 3D fast gradient echo spin warp sequence with identical imaging parameters.
Discussion Because MR projection imaging involves parallel rays, the full resolution of a complete angular scan is produced everywhere in the FOV. Use of limited angle acquisition provides adequate information for the reconstruction of a limited FOV provided that artifacts from objects from a distance larger than the radius of the local small FOV are inconsequential. Within the limitations of this restriction, each point in the image enjoys benefits similar to those of small FOV spin warp imaging.
Results Figure 2 compares images of a resolution phantom acquired with spin warp and PR. The top images show the entire phantom. The area indicated by the black box (upper left image) has been magnified in the bottom images. The images on the left are a 512 x 128 spin warp acquisition. The phase encoding direction is horizontal. The images on the right are a 512 X 128 angle PR. The horizontal dots are seen better in the PR image, which has resolution comparable to a 512 x 512 spin warp reconstruction.
Conclusions Limited angle PR imaging has the potential for substantially improved resolution or speed in applications where artifacts generated by objects outside the effective local FOV can be tolerated. References 1. J. Pauly D. Nishimura, A. Macovski, J. Magn. Res. 81, 43, 1989. 2. J. 0. Fredrickson, N.J. Pelc, MRM 35, 621, 1996. 3. D. Noll, et al. SPIE, Medical Imaging V, 1443, 29, 1991. 4. V. Rasche, D. Holz, W. Schepper, MRM 32, 629,1994. 5. M.R. Smith., T.M. Peters, R.H.T.Bates, J Phys. A. 6, 361, 1973
