Quantification of the PET ACR phantom using the GE SIGNA PET/MRI system

Objectives Attenuation correction of patient data on the GE SIGNA PET/MR system utilizes the LAVA FLEX pulse sequence to separate fat and water. This technique however, results in large errors for phantom studies primarily due to the lack of any fat content in the phantom as shown in previous studies [1]. In this regard, several attenuation map templates corresponding to different phantoms (such as the ACR, IEC, well counter calibration) have been created to correct for attenuation when these phantoms are imaged. However, such a process does not test the accuracy of attenuation correction of patient studies due to the difference in the two approaches. In this work we investigate ways for the LAVA FLEX technique to correctly identify the various components of a phantom - with specific focus on the ACR phantom - to result in accurate PET attenuation without relying on preset templates.

Methods An ACR phantom with a modified lid consisting of 5 x 25mm inner diameter cylinders - two with air, one with F-18 solution, and two with cold water - was used. The phantom was scanned and reconstructed on a GE 710 PET/CT and on a GE MR750. AC maps of the phantom were derived from CTAC and from LAVA FLEX images. MR images of the phantom were acquired when the two cold water cylinders were filled with tap water, and when one was filled with vegetable oil to allow for water and fat separation. Two PseudoCTs(PCT) were created from the MR LAVA FLEX images of these phantoms using the GE Signa PET/MR abdominal processing station which assigns Houndsfield Units(HU) for the following material classes: air outside the subject(-1000), continuous fat/soft tissue values between -104 and 42, and air pockets inside the subject(-1000). Two derivative PCTs(dPCT) from the scan with vegetable oil were also made, one with the HU values between soft tissue and water rescaled with a maximum HU value of water, and one in which a digitally added 5mm acrylic shell with HU of 120 was added to the previous dPCT. The patient table from the CTAC, was digitally inserted into the PCT and dPCTs. Reconstructions of the PET data acquired on the PET/CT were performed using the two PCTs and two dPCTs. Phantom filling, activity levels, imaging, and analysis was performed according to ACR requirements using an injection of 370 MBq For each reconstruction, a quantitative comparison for the background and cylinders was made with respect to CTAC as the gold standard.

Results Visually, there were no differences in spatial resolution and uniformity scores of the cold rods and uniform section of the phantom between the different reconstructions. With no oil in one of the cylinders, the phantom background was improperly assigned HU of fat, however with oil in one of the cylinders, the phantom background was assigned HU corresponding to soft tissue. The Air Cylinder assigned HU values were between fat and water. Quantitative results for background, hot cylinder, and air are tabulated. Values in parenthesis represent percent error.

Conclusions With oil added to one of the cylinders, the AC map has proper fat/water separation, however, over attenuation occurs as an HU of 42 is assigned to water and -100 is assigned to the air cylinders. We demonstrated that rescaling soft tissue HU values to water, and digitally adding acrylic to account for the phantom shell, results in an SUVmean background error of -2%, and a Hot Cylinder SUVmax difference of -4%. Future clinical MR AC protocols might include Zero Echo Time pulse sequences which could potentially image the acrylic shell and internal acrylic structures, eliminating the need to digitally account for the phantom acrylic shells.