TERSE: a transmission and emission reconstruction environment for SPECT, has been developed at the University of Michigan by Jeff Fessler and Edward Ficaro, in a collaboration between the Department of Electrical Engineering and Computer Science (EECS) and the Division of Nuclear Medicine.
TERSE consists of an X-windows graphical user interface to portions of the ASPIRE software libraries that have been optimized for SPECT imaging with simultaneous transmission and emission scans. Software components include: data structure for storing sparse matrices; subroutines for linear algebra operations with sparse matrices; programs for generating the “system matrices” for a wide variety of tomographic systems including parallel beam, fan beam, depth-dependent blurring, etc.; fast converging iterative algorithms for computing penalized likelihood and penalized weighted least squares estimates from noisy data; a variety of convex penalty functions are implemented.
TERSE has been adopted for routine use in cardiac scans at the University of Michigan Medical Center, and its first years of use over 2000 cardiac scans were reconstructed using this method. We initially used a Picker PRISM 3000 triple-head SPECT system for cardiac scans, with an Americium transmission source developed at UM opposing a 65-cm fan-beam collimator, with parallel-beam collimators on the other two heads. All three heads collection emission data. An iterative coordinate ascent algorithm in the TERSE package produces cross-sectional thorax attenuation images from the fan-beam transmission data in which the lungs, spine, and soft-tissue are clearly seen. The ring artifact that is intrinsic to conventional FBP images from truncated measurements is eliminated by our statistical approach. We incorporate these nonuniform attenuation characteristics into an emission reconstruction algorithm, which yields images with improved uniformity of the cardiac wall in normal patients, thus reducing the false positives (which lead to unnecessary risky and expensive catheterization procedures).
Unlike commercially available iterative reconstruction methods, TERSE is based on regularized statistical methods, and uses algorithms with guaranteed global convergence. This means the image properties, such as noise and spatial resolution, are more predictable and controllable than with the popular unregularized methods. The algorithms converge quickly to a stable image. The trade-off between resolution and noise is controlled directly by the user, not indirectly by the number of iterations, number of counts, etc.
TERSE has been evaluated by ROC studies using both human observers and semi-automatic polar-map based observers, and statistically significant improvements in diagnostic accuracy have been demonstrated over conventional FBP without attenuation correction. See 1996 Circulation paper: “Simultaneous transmission/emission myocardial perfusion tomography: Diagnostic accuracy of attenuation-corrected 99m-Tc-Sestamibi SPECT” by Ficaro et al. A full reference to the paper is available in the publication list at http://www.eecs.umich.edu/~fessler/papers.)
Representative images reconstructed with the TERSE software can be viewed at http://www.eecs.umich.edu/~fessler/result/terse.