Evaluation of sinogram consistency conditions identifying artifactual PET attenuation images

In PET/CT, PET attenuation maps are produced from the CT image resulting in localized defects due ambiguities in the conversion process. An example of a source of difficulty is CT contrast agent. The suitability of sinogram consistency conditions (SCCs) for objective functions (OFs) in algorithms for assessing the presence of artifacts is evaluated. Given a measured sinogram and two possible attenuation images the accuracy with which SCC-based OFs identify the artifact-free image is gauged. Digital phantom sets, each consisting of a true emission and attenuation component, were used. From each, 128 2D "measured" sinogram statistical variates at 7 noise levels representing sinogram counts from 10³ to 10⁷ were produced. Artifactual attenuation images were made by modifying the true attenuation phantoms. True and artifactual attenuation corrected sinograms (test pair) for each measured sinogram were produced. SCCs state that with consistent data, certain "forbidden" coefficients (FCs) from the Fourier transform over angle of the various radial moments of the corrected sinogram vanish. Schemes for combining FCs into an OF were investigated and classification accuracy was determined. A high-performing SCC-based OF produced an accuracy of 92% at the 10⁷ count level for phantoms with a 2 cm² artifact. Performance fell with increasing noise. At 10⁵counts accuracy was 53% (chance=50%). Better performance was obtained with larger artifacts. SCC objective functions may be useful in artifact correction, but are limited by low-count-level performance