Attenuation correction in PET/CT from ultra-low dose CT: Photon starvation and iterative CT reconstruction

Objectives With the current generation of PET/CT scanners, CT scan data must be used for attenuation correction. Even with a low dose technique, this contributes substantially to the total effective dose. Using low or ultra-low dose CT techniques ameliorates this but may introduce substantial artifacts, particularly those due to photon starvation in bone-rich cross-sections. The objective of this work was to characterize the attenuation-correction errors resulting from photon starvation and explore whether an iterative CT reconstruction algorithm would reduce them.

Methods Data were acquired using a GE DLS PET/CT scanner and a GE CT750 HD scanner equipped with ASiR iterative reconstruction. A Data Spectrum torso phantom was scanned with a variety of CT techniques and then processed to produce 511 keV attenuation maps. Slabs of material consisting of a mix of copper and aluminum, were attached to the periphery of the phantom to simulate additional bone.

Results CT data introduced errors into the PET attenuation correction map in two ways: 1) attenuation coefficients derived from noisy CT data are biased due to the non-linear mapping of Hounsfield units (HU) to 511 keV (mu-511) attenuation coefficients; and 2) photon starvation artifact directly produces a bias in HU measured from the CT data and thus errors in mu-511. The latter effect was most dominant for ultra-low dose CT techniques, particularly for lower kVp, and increased with the amount of bone. ASiR (100%) reconstructions reduced the image noise (CT noise index by a factor of approximately two) but had no effect on bias due to photon starvation.

Conclusions Photon starvation artifact in cross-sections containing large amounts of dense bone affect the CT-based attenuation correction leading to substantial errors in SUV measurements. Reconstructing CT attenuation scans with the ASiR (100%) algorithm reduced noise but did nothing to reduce the bias in attenuation estimation