http://arxiv.org/abs/1702.04993
-In the Alvarez-Macovski method [R.E. Alvarez and A. Macovski, Phys. Med. Biol., 1976, 733-744], the attenuation coefficient is approximated as a linear combination of functions of energy multiplied by coefficients that depend on the material composition at points within the object. The method then computes the line integrals of the basis set coefficient from measurements with different x-ray spectra. This paper shows that the transformation from photon counting detector data with pileup to the line integrals can become ill-conditioned under some circumstances leading to highly increased noise. Methods: An idealized model that includes pileup and quantum noise is used. The noise variance of the line integral estimates is computed using the Cram{\`e}r-Rao lower bound (CRLB). The CRLB is computed as a function of object thickness for photon counting detector data with three and four bin pulse height analysis (PHA) and low and high pileup. Results: With four bin PHA data the transformation is well conditioned with either high or low pileup. With three bin PHA and high pileup, the transformation becomes ill-conditioned for specific values of object attenuation. At these values the CRLB variance increases by approximately 10 5 compared with the four bin PHA or low pileup results. The condition number of the forward transformation matrix also shows a spike at those attenuation values. Conclusion: Designers of systems using counting detectors should study the stability of the line integral estimator output with their data.
R. Alvarez
Fri, 17 Feb 17
42/43
Comments: arXiv admin note: text overlap with arXiv:1702.01006