Evaluation of emission-transmission registration in thoracic PET

The intent of this investigation was to quantitate the amount of misregistration between PET emission and transmission scans of the thorax that occurs in a normal clinical environment.

The data from 17 FDG myocardial studies were evaluated. Prior to injection, a transmission study was acquired for 15 min using a 68Ge/68Ga ring source. The location of the cross-hairs from a laser alignment system was marked on the patient who was then removed from the scanner and injected with 10 mCi of FDG. After 45 min, the patient was placed back on the table and repositioned with the previously placed marks and a 15-min emission scan was acquired. The outline of the lungs on both the transmission and emission images was manually segmented. Both attenuation-corrected and noncorrected emission images were evaluated and the one that provided clearer visualization of the outline of the lungs was chosen for segmentation. The segmented contours of the transmission and emission scans were then registered with the method described by Pelizzari et al. using the transmission image as the "head" and the emission image as the "hat." The allowable transformations were x and y shifts and rotation in the transverse plane.

Shifts in the x-axis averaged 2.4 mm (range: 0.2-7.3 mm, 80% less than 3.3 mm) with shifts in the y-axis averaging 2.6 mm (range: 0.1-8.7 mm, 80% less than 2.4 mm) and rotations in the transverse plane averaging 1.6 degrees (range: 0.2 to 5.1 degrees, 80% less than 2.4 degrees). A phantom study indicated that the accuracy of this method of evaluating misregistration was 2.35 mm and 1.81 mm in the x and y directions, respectively.

Our preliminary evaluation indicates that careful application of laser alignment is an adequate method of registration in most cases.