Performance Characteristics Obtained for a New 3-Dimensional Lutetium Oxyorthosilicate–Based Whole-Body PET/CT Scanner with the National Electrical Manufacturers Association NU 2-2001 Standard

Source Preparation and Acquisition Protocol.

The N-01 standard specifies that the activity should be placed for an axial extent of less than 1 mm in a glass capillary tube with an ID of <1 mm. It is further required that the source be placed parallel to the tomograph long axis for all measurements. As pointed out by Erdi et al. (2), positron annihilation is not confined to the location of ¹⁸F in the fluid; positrons may stream up the capillary and annihilate in the glass of the tube. This positron range effect may tend to elongate the imaged source in the direction of the tube; indeed, Erdi et al. measured a significantly wider axial resolution when the source was placed along the axial direction than when it was placed transaxially in the scanner, demonstrating that the source was asymmetric. Consequently, we used the method suggested by Bailey et al. (7) for the point source fabrication; we used commercially available molecular sieves (Zeolites; Molecular Sieve 13X, 8/12 beads 100g, part 05268 [Alltech Associates]) for adsorption of radioactivity onto small (2-mm-diameter) beads. Soaking the beads in an ¹⁸F-FDG solution of ∼0.5 GBq/mL for 1–2 min, we produced sources containing 2.5–3 MBq each. The total activity placed in the FOV was thus <10 MBq for each acquisition. Under these conditions, as demonstrated by Watson et al. (6), the N-01 criterion that both dead time and the ratio of random events to total events be <5% was fulfilled. The point sources were placed on a piece of tape, which was suspended by means of a needle. By use of a source holder, the 3 point sources were positioned in the scanner FOV as follows: x = 0 cm, y = 1 cm; x = 0 cm, y = 10 cm; and x = 10 cm, y = 0 cm. Once in place, the 3 point sources were aligned (axially) in the scanner FOV by use of laser lights. Two sets of emission measurements were obtained with the sources centered at 2 axial positions in the scanner FOV: in the center and at one quarter of the axial FOV (4.05 cm). For each position, more than 2 × 10⁶ counts were acquired to ensure adequate statistics.

Reconstruction and Data Analysis.

Data were corrected for scanner geometry, crystal efficiency, random coincidences (by use of the delayed coincidence window technique), and system dead time. For each position, the images were reconstructed by use of the filtered backprojection (FBP) algorithm onto a 336 × 336 matrix with a ramp filter. The reconstruction zoom was set to 2. The pixel size was 1 mm. Transverse spatial resolution was calculated for each point source position as full width at half maximum (FWHM) and full width at 10th maximum (FWTM) of the resulting point-spread function by interpolating the adjacent pixels on the radial and tangential profiles. An axial profile was derived from the number of counts in each slice versus the slice number, and axial resolution was measured as the FWHM and FWTM of such a profile. Radial and tangential resolutions (FWHM and FWTM) for each radial position (1 and 10 cm) were averaged for both of the axial positions.

Source Preparation and Acquisition Protocol.

A polyethylene tube (ID, 1 mm; OD, 3 mm; length, 70 mm) was filled with a total of 2.14 MBq and placed in the center of 5 concentric aluminum sleeves (IDs, 3.9, 7.0, 10.2, 13.4, and 16.6 mm). The prepared phantom was suspended in the center of the transaxial FOV and oriented parallel to the scanner long axis. To compensate for counting-rate losses as the source decayed, the measurements were started with all of the aluminum tubes in the phantom. After each acquisition (240 s each), the most external tube was removed, and the acquisition was repeated. The same measurements were repeated at a 10-cm radial offset from the center of the transaxial FOV.

Data Analysis.

Raw-data sinograms were used for sensitivity analysis. Single-slice rebinning was applied to the sinograms to assign counts in the oblique line of responses to the image slices where the line of responses crosses the scanner axis. The total system sensitivity at each radial position was calculated by dividing the total counting rate in the absence of any attenuating material (R₀) by the corresponding activity. R₀ was determined by fitting the natural logarithm of measured counting rates, after correction for activity decay, as a function of the sleeve thickness. Linear regression was used to fit the data, obtaining an extrapolated value for R₀. The sensitivity for each plane was calculated according to the N-01 standard.

Source Preparation and Acquisition Protocol.

The phantom used was the 20-cm-diameter solid polyethylene cylinder with an overall length of 70 cm and a line source threaded along its length, parallel to its central axis, at 4.5 cm from its center. The phantom was centered both in the transverse FOV and in the axial FOV so that the line source was at the position nearest the patient table. The line source was filled with 1,047.47 MBq of ¹⁸F initial activity, enough to achieve counting rates beyond the peak trues and NEC rates. Forty-six frames were acquired: the first 10 with a 10-min duration, with a 5-min gap between the scans, during which the phantom was extracted from the FOV, the second 10 with a 15-min duration, and the last 26 with a 20-min duration, with a 10-min gap between the scans. Data were acquired over 18 h, and for each acquisition, more than 2.5 × 10⁶ coincidence counts were stored. Separate prompt and delayed sinograms were acquired by use of the standard delayed coincidence window technique.

Data Analysis.

3D sinograms were rebinned by use of the single-slice rebinning algorithm. Data analysis was carried out by the modified procedure suggested by Watson et al. (6). Instead of estimating the randoms (R_randoms) from the prompt sinograms by correcting the totals (R_TOT) for trues (R_trues) and scatter (R_scatter), these values were computed directly from the delayed window sinogram with the same bins as those used for the prompt sinogram analysis. The scatter event rate for each slice (i) at each activity (j) was computed as follows: neglecting the contribution of intrinsic trues. The SF was estimated as follows: NEC rates were calculated as follows: Two NEC rate curves were generated, for k = 1 (noiseless random correction) and k = 2 (noisy random correction).

Source Preparation and Acquisition Protocol.

The same phantom as that used for the SF and NEC rate tests was used. The line source was filled with 635 MBq of ¹⁸F initial activity, enough to achieve counting rates beyond the peak NEC rates (k = 2). Twenty-six frames were acquired: the first 11 with a 15-min duration, the intermediate 11 with a 20-min duration, followed by a 40-min gap between the scans, and the last 4 with a 20-min duration. Data were acquired over 14.75 h, and for each acquisition, more than 2.5 × 10⁶ coincidence counts were stored.

Reconstruction and Data Analysis.

All counting-rate–dependent corrections (dead-time losses and random coincidences) were applied. Data were reconstructed by use of the Fourier rebinning–ordered-subset expectation maximization (OSEM) iterative reconstruction, with 2 iterations and 8 subsets. A 4-mm FWHM gaussian filter was applied to the image after reconstruction along the axial and transaxial directions. The data were reconstructed over a 128 × 128 matrix with a 2-mm pixel size and slice thickness.

Once reconstructed, the images were analyzed by drawing a circular region of interest (ROI) (18-cm diameter) centered on the reconstructed images of the phantom. The counting-rate error (%ΔR) as a function of the effective activity concentration was calculated as the deviation of the true counting rate (R) from a linear trend extrapolated from the low-activity acquisitions (6), at which dead-time losses were negligible (R_extrap), as follows:

Source Preparation and Acquisition Protocol.

Image quality was evaluated by imaging a torso phantom containing 6 coaxial isocentered spheres with IDs of 10, 13, 17, 22, 28, and 37 mm. A 5-cm-diameter cylindric insert filled with foam pellets with an average density of 0.30 g/mL was positioned in the center of the phantom to simulate lung tissue. The phantom was filled with a solution of water and ¹⁸F (5.4 kBq/mL) and the 4 smallest spheres with a radioactivity concentration 7.9 times that of the background (lesion-to-background activity ratio [L/B] = 8). The remaining 2 spheres (28- and 37-mm IDs) were filled with nonradioactive water. In a second acquisition, the radioactivity concentration in the hot sphere was 3.9 times that of the background (L/B = 4). Once filled, the phantom was positioned centrally in the scanner FOV (both axially and transaxially). To simulate the body activity from outside the scanner FOV, the 70-cm scatter phantom was positioned at the edge of the torso phantom. The plastic tube inside the scatter phantom was filled with 119,2 and 112,2 MBq of ¹⁸F for the 2 acquisitions. The imaging time was set to simulate a total body scan (100-cm total axial imaging distance in 60 min, according to the N-01 standard). Because the transmission scan is performed with the CT scanner, the contribution of its relative scan time to the total imaging time can be considered negligible. Data acquisition time was determined by considering the 12.2-cm (16.2- to 4-cm overlap) axial distance between consecutive bed positions in a total-body scan to be 7.32 min. We used 0.32 min for the positioning and CT scans and 7 min for the PET emission scan.

Reconstruction and Data Analysis.

Data were corrected for random coincidences, normalization, dead-time losses, scatter, and attenuation. The results are reported for 3 different reconstructions: 1 FBP reconstruction (all-pass filter) and 2 attenuation-weighted OSEM iterative reconstructions, with 2 iterations and 8 subsets and with 4 iterations and 14 subsets. In all cases, Fourier rebinning was used to reduce the 3D dataset to a 2-dimensional–equivalent one, and a 4-mm FWHM gaussian filter was applied to the image after reconstruction along the axial and transaxial directions. The data were reconstructed over a 128 × 128 matrix with a 5.25-mm pixel size and a 2-mm slice thickness.

To evaluate the hot- and cold-sphere contrast, circular ROIs with diameters equal to the physical IDs of the spheres were drawn over PET images. Twelve background ROIs (37-mm diameter) were drawn on the central slice as well as on slices at ±10 mm and ±20 mm axially. Smaller ROIs (10, 13, 17, and 28 mm) were drawn concentric to the 37-mm background ROIs. In addition, a 5-cm ROI was drawn (in each phantom slice) on the central lung insert to assess the accuracies of the attenuation and scatter corrections. To minimize the influence of the operator in the ROI positioning, an automatic routine was developed with IDL 6.1 (Research Systems, Inc.); the operator needed only to establish the center of the 13-mm sphere and of the lung insert in the central slice. Positioning and determination of dimensions of the remaining ROIs on spheres, the background, and the lung insert were done automatically and closely following N-01 recommendations. ROI analysis was performed by taking into account partial pixels as well.

The hot-sphere contrast recovery coefficient (Q_H) was calculated as follows: where C_hot and C_bkgd are the averages of the counts measured in the hot-sphere ROI and of the counts measured in all background ROIs, respectively, and a_hot/a_bkgd is the ratio of activities in the hot sphere and in the background.

The cold-sphere contrast recovery coefficient (Q_C) was calculated as follows: where C_cold is the average of the counts measured in the cold-sphere ROI.

The accuracies of the attenuation and scatter corrections were calculated as the residual errors in the lung region as follows: where C_lung is the average counts in the lung ROI. The variation coefficients of the means (N_j) in the background ROIs for the various sphere sizes were taken as a measure of background variability, as follows: where SD_j is the SD of the background ROI counts for sphere j.

To assess the effect of image noise level on contrast measurement, repeated scans over a range of activity levels for L/B = 8 and L/B = 4 activity ratios were performed. The measured activity levels were 1.7, 1, and 0.59 times the standard background activity level. For this study, only FBP reconstruction was used. The percent variation coefficients of the contrast values across the range of activity levels were computed.