Clinical evaluation of processing techniques for attenuation correction with 137Cs in whole-body PET imaging

Transmission scanning can be successfully performed with a 137Cs single-photon emitting point source for three-dimensional PET imaging. However, the attenuation coefficients provided by this method are underestimated because of the energy difference between 662- and 511-keV photons, as well as scatter and emission contamination when the transmission data are acquired after injection. The purpose of this study was to evaluate, from a clinical perspective, the relative benefits of various processing schemes to resolve these issues.

Thirty-eight whole-body PET studies acquired with postinjection singles transmission scans were analyzed. The transmission images were processed and applied to the emission data for attenuation correction. Three processing techniques were compared: simple segmentation (SEG) of the transmission scan, emission contamination subtraction with scaling (ECS) of the resulting data to 511-keV attenuation coefficient values and a hybrid technique performing partial segmentation of some tissue densities on the ECS scan (THR). The corrected emission scans were blindly assessed for image noise, the presence of edge artifacts at the lung-soft-tissue interface and for overall diagnostic confidence using a semiquantitative scoring system. The count densities and the SDs in uniform structures were compared among the various techniques. The observations for each method were compared using a paired t test.

The SEG technique produced images that were visually less noisy than the ECS method (P < 0.0001) and the THR technique, but at the expense of increased edge artifacts at the boundaries between the lungs and surrounding tissues. The THR technique failed to eliminate these artifacts compared with the ECS technique (P < 0.0001) but preserved the activity gradients in the hilar areas. The count densities (and thus, the standardized uptake values) were similar among the three techniques, but the SEG method tended to underestimate the activity in the lung fields and in chest tumors (slope = 0.79 and 0.94, respectively).

For many clinical applications, SEG data remain an efficient method for processing 137Cs transmission scans. The ECS method produced noisier images than the other two techniques but did not introduce artifacts at the lung boundaries. The THR technique, more versatile in complex anatomic areas, allowed good preservation of density gradients in the lungs.