Recovery of localized defects in PET attenuation-correction images

CT data from PET/CT can be used to produce a 511-keV attenuation map of high statistical quality for PET attenuation correction. However, CT contrast agents can result in regions of overestimated 511 keV attenuation-coefficient values. Resulting attenuation corrected emission images have been observed with focal artifacts resembling diseased tissue. We investigate a method for recovering the appropriate values of artifactual attenuation by using the measured emission data. The recovery problem is simplified by the assumption of an otherwise perfect attenuation map containing only limited regions of incorrectly assigned attenuation. Additionally a "binary" constraint may be applicable to the recovery of contrast induced artifacts in which it is assumed that a suspect pixel with a high attenuation coefficient value is either (a) correct or (b) misidentified contrast that should have the value of water. A demonstration recovery algorithm is introduced. To start, suspected regions within the attenuation map are defined. Such a region could be, for example, all pixels with high attenuation values. An iterative procedure is described in which updates of the attenuation and emission images are interleaved. In the attenuation case only pixel values within the defined regions are changed. The binary constraint is incorporated into the procedure. Results with several classes of digital phantoms yield perfect recovery in the noise free case. Stochastic noise degrades algorithm performance.