Impacts of acquisition and reconstruction parameters on the absolute technetium quantification of the cadmium–zinc–telluride-based SPECT/CT system: a phantom study

Background The digital cadmium–zinc–telluride (CZT)-based SPECT system has many advantages, including better spatial and energy resolution. However, the impacts of different acquisition and reconstruction parameters on CZT SPECT quantification might still need to be validated. This study aimed to evaluate the impacts of acquisition parameters (the main energy window and acquisition time per frame) and reconstruction parameters (the number of iterations, subsets in iterative reconstruction, post-filter, and image correction methods) on the technetium quantification of CZT SPECT/CT. Methods A phantom (PET NEMA/IEC image quality, USA) was filled with four target-to-background (T/B) ratios (32:1, 16:1, 8:1, and 4:1) of technetium. Mean uptake values (the calculated mean concentrations for spheres) were measured to evaluate the recovery coefficient (RC) changes under different acquisition and reconstruction parameters. The corresponding standard deviations of mean uptake values were also measured to evaluate the quantification error. Image quality was evaluated using the National Electrical Manufacturers Association (NEMA) NU 2–2012 standard. Results For all T/B ratios, significant correlations were found between iterations and RCs (r = 0.62–0.96 for 1–35 iterations, r = 0.94–0.99 for 35–90 iterations) as well as between the full width at half maximum (FWHM) of the Gaussian filter and RCs (r = − 0.86 to − 1.00, all P values < 0.05). The regression coefficients of 1–35 iterations were higher than those of 35–90 iterations (0.51–1.60 vs. 0.02–0.19). RCs calculated with AC (attenuation correction) + SC (scatter correction) + RR (resolution recovery correction) combination were more accurate (53.82–106.70%) than those calculated with other combinations (all P values < 0.05). No significant statistical differences (all P values > 0.05) were found between the 15% and 20% energy windows except for the 32:1 T/B ratio (P value = 0.023) or between the 10 s/frame and 120 s/frame acquisition times except for the 4:1 T/B ratio (P value = 0.015) in terms of RCs. Conclusions CZT-SPECT/CT of technetium resulted in good quantification accuracy. The favourable acquisition parameters might be a 15% energy window and 40 s/frame of acquisition time. The favourable reconstruction parameters might be 35 iterations, 20 subsets, the AC + SC + RR correction combination, and no filter. Supplementary Information The online version contains supplementary material available at 10.1186/s40658-021-00412-4.

energy window subtraction method [23,24], energy-weighted acquisition method [25,26], inverse Monte Carlo reconstruction algorithm, and so on [27,28]. Attenuation correction (AC) has been achieved by using the CT-based attenuation correction method [29], Chang algorithm method [30], and so on. Additionally, some methods have been applied to reduce the partial volume effect, including image enhancement techniques [31,32], image domain anatomically-based PVC (partial volume correction) techniques [33], projection-based PVC, and so on [34]. Despite this, both the spatial and energy resolutions of conventional NaI SPECT are relatively low, and radionuclides applied in SPECT have a higher fraction of scattered photons than those of PET [35,36]. These drawbacks might magnify the partial volume effect and compromise the effectiveness of SC and AC. As a result, SPECT images may be more difficult to quantify.
Today, however, with the development of SPECT systems, absolute quantification is also widely validated and used. Some studies have suggested that absolute SPECT quantification is promising with different SPECT equipment when reconstruction protocols are standardized [37,38]. SPECT quantification of various radionuclides has also been well studied, including technetium-99 m ( 99m Tc) [39], indium-111 ( 111 In) [40], yttrium-90 ( 90 Y) [41], lutetium-177 ( 177 Lu) and so on [42]. Additionally, SPECT quantification has been widely used in clinical practice, such as quantification of the lung shunt fraction in hepatic radioembolization [43], myocardial perfusion imaging [44], monitoring cancer [45], and determining lesion volumes [46]. Despite these validations and clinical practices, various acquisition and reconstruction parameters may also affect the accuracy of SPECT quantification. Some studies have suggested that the small number of iterations and subsets used in OSEM (ordered subsets expectation maximization) reconstruction influences the quantification accuracy because of incomplete convergence [47,48]. The application of different correction methods, such as AC, SC, and RR (resolution recovery correction), may also affect the quantification accuracy [29,49,50]. However, many of these studies are based on conventional NaI SPECT systems, and therefore, the impacts of different acquisition and reconstruction parameters on absolute CZT SPECT quantification might need to be studied. This study aimed to evaluate the impacts of acquisition parameters (the main acquisition energy window and acquisition time/frame) and reconstruction parameters (the number of iterations and subsets in iterative reconstruction, post-filter, AC, SC, and RR correction) on the accuracy of CZT SPECT/CT technetium quantification.

Image acquisition parameters
SPECT/CT acquisition of the PET NEMA/IEC image quality phantom was performed on a Discovery NM/CT 670 CZT (GE Healthcare, USA) equipped with wide energy high-resolution collimators. All SPECT images were acquired with a list mode. The step and shoot acquisition mode was performed by 360-degree rotations (120 s/6degree per frame) with a matrix size of 128 × 128 without zoom. Two main energy windows (140 keV ± 7.5% and 140 keV ± 10%) were reconstructed by the list mode to evaluate the impacts of the main energy window on RCs. The scatter energy window was 120 ± 5% keV. CT images were acquired with a 120 kVp tube voltage, 200 mA tube current, matrix size of 512 × 512, and 1.25 mm slice thickness.

Image reconstruction parameters
All images were reconstructed using the OSEM algorithm with 1-90 iterations and 2-30 subsets [52]. The FWHM range of the Gaussian filter was 0.7-6.99 mm. The correction methods used in this study included CT-based AC, dual-energy-window technique-based SC, and point spread function-based RR correction. Three image correction combinations were used to evaluate the impacts of the image correction methods, including AC + SC + RR, AC + SC, and AC + RR. List mode was applied to reconstruct the acquisition time to 1-120 s/frame. In every step of the analysis, we evaluated the impact of a certain parameter to determine the optimal value while fixing all other parameters at the same time. All acquisition and reconstruction parameters in the evaluation process are listed in Fig. 1. Fig. 1 Analysis process of different acquisition and reconstruction parameters. The process started with six subsets, no filter, AC + SC + RR correction combination, 120 s/frame of acquisition time, and 15% energy window. The impacts of iterations, subsets, FWHM, correction combination, energy window, and acquisition time/frame were evaluated in sequence and an optimal value of these parameters was determined in each step of the process

RC calculations
Volumes of interest (VOIs) of six spheres were delineated using the inner edge of spheres of CT images as references. Mean uptake values (MBq/ml) and corresponding standard deviations (SDs) were automatically calculated three times by the Q. Metrix of GE-Xeleris 4.0 workstation (GE Healthcare, USA) and are shown as averages. RCs were calculated using Eq. (1) [19]:

Image quality evaluations
To assess the image quality, we calculated both the per cent contrast and coefficient of variation (COV) complying with the NEMA NU 2-2012 standard [53][54][55][56]. The per cent contrast Q H,j for each hot sphere was calculated by using Eq. (2): where C H,j is the average count in the region of interest (ROI) for sphere j, C B,j is the average of the background ROI counts for sphere j, a H is the activity concentration in the hot spheres, and a B is the activity concentration in the background. The COV N j for each hot sphere was calculated by using Eq. (3): where SD j is the standard deviation of the background ROI counts for sphere j.

Statistical analysis
All statistical analyses were performed by SPSS 23.0 (IBM, USA). All graphs were produced by GraphPad Prism 8.3.0 (GraphPad Software, USA) and Origin Pro 2021 (OriginLab, USA). The relationships between RCs and the different number of iterations and subsets, FWHM, and acquisition time/frame were established by Pearson's rank correlation and linear regression analysis. RCs and per cent contrast of three different correction combinations were compared using the paired t-test [57]. The comparison of RCs for different energy windows and acquisition time/frame was also analysed by using the paired t-test. A P value lower than 0.05 was considered statistically significant. Figure 2 shows that of all the T/B ratios, the RCs of larger spheres converged earlier than those of smaller spheres, of which 37-17 mm spheres converged at 35 iterations and 13-10 mm spheres converged at 85 iterations. Table 1 shows that apart from the 37 mm sphere of 1-35 iterations of the 32:1 and 16:1 T/B ratios, there were significant positive (1) RC = Mean measured radioactivity concentration Actual radioactivity concentration × 100%.  Figure 3 shows that RCs did not increase rapidly with an increasing number of subsets. RCs of larger spheres (37-17 mm) became stable after 20 subsets. Table 1 shows that the Pearson r values of the six spheres ranged from 0.68 to 0.89 (32:1 T/B ratio), 0.61 to 0.90 (16:1 T/B ratio), 0.52 to 0.85 (8:1 T/B ratio), and − 0.34 to 0.98 (4:1 T/B ratio). In linear regression analysis, the regression coefficients of the six spheres ranged from 0.09 to 0.70 for the 32:1 T/B ratio, 0.07 to 1.14 for the 16:1 T/B ratio, 0.08 to 0.90 for the 8:1 T/B ratio, and − 0.06 to 1.14 for the 4:1 T/B ratio.

Impacts of the Gaussian filter
In Fig. 4, RCs of all spheres declined significantly as the FWHM (0.7-6.99 mm) of the Gaussian filter increased. There were significant negative correlations between the

Impacts of the image correction methods
The profiles of four T/B ratios show that RCs of the AC + SC + RR correction combination were closer to the actual sphere activity concentration. The AC + RR combination predicted the highest mean uptake values, while the AC + SC combination predicted the lowest mean uptake values in spheres (Fig. 5). Table 2 shows that the RCs of the AC + SC + RR combination were lower than those of the AC + RR combination but higher than those of the AC + SC combination ( Figure 6 shows the visual difference of the images reconstructed with different correction combinations. Among all T/B ratios, the AC + SC + RR combination had a better visual image quality. Table 3 shows that the per cent contrasts of six spheres reconstructed using the AC + SC + RR combination were higher than those of other correction combinations (AC + SC + RR vs. AC + RR; AC + SC + RR vs. AC + SC, all P values < 0.05). However, the COVs of the AC + RR combination were lower than those of the AC + SC + RR combination or AC + RR combination (all P values < 0.05). The COVs of the AC + SC combination were higher than those of the AC + SC + RR combination (100. 70 Table 4).

Impacts of the main energy window
As shown in Table 5, for the 32:1 T/B ratio, RCs under the 15% energy window were higher than those under the 20% energy window, and there was a statistically significant difference (P value = 0.023). However, for lower T/B ratios (16:1, 8:1, and 4:1), there were no statistically significant differences (all P values > 0.05). Since a 15% energy window might improve the quantification accuracy for a higher T/B ratio (32:1), it was determined to be the optimal energy window in this step.

Impacts of the acquisition time per frame
The correlations between RCs and the acquisition time/frame were − 0.

Discussion
In this study, CT images were applied as references to avoid the partial volume effect of the SPECT system to calculate RCs with lower errors [58]. The study of Dr. Koole, M et al. suggested that high-resolution structural information from MR or CT images is helpful in determining potential lesions in SPECT images [59]. This study showed that the number of iterations had a large impact on quantification. Figure 2 shows that RCs of larger spheres (37-17 mm) converged earlier than those of smaller spheres (13 mm and 10 mm spheres). This indicated that a small number of iterations might be enough for larger lesions in absolute quantification. Although the correlations between RCs and 1-35 iterations were lower than those of 35-90 iterations, 1-35 iterations had much higher regression coefficients than those of 35-90 iterations ( Table 1). RCs could also increase rapidly within the first 35 iterations for all spheres (Fig. 2). This indicated that although 35-90 iterations had strong linearity, it could not increase RCs efficiently because of the much smaller regression coefficients compared with those of 1-35 iterations. Furthermore, smaller spheres had relatively larger errors, and more iterations could increase RCs, but not efficiently. Therefore, it was determined that the optimal number of iterations might be 35.
The correlations between subsets and RCs were not obvious because, for many spheres, the P values were greater than 0.05 (Table 1). Additionally, the regression coefficients of all T/B ratios were very low (0.09-0.70 for the 32:1 T/B ratio, 0.07-1.14 for the 16:1 T/B ratio, 0.08-0.90 for the 8:1 T/B ratio, and − 0.06 to 1.14 for the 4:1 T/B ratio) ( Table 1). These results indicated that RCs could not increase rapidly with the increasing number of subsets; therefore, subsets had a relatively small impact on quantification. The study of Dr. Vriens, D et al. also suggested that subsets have only a small effect on the standardized uptake value (SUV) in phantom experiments [60]. In this study, RCs tended to be stable after 20 subsets for the larger spheres (37-17 mm) and did not increase significantly after 20 subsets for the smaller spheres (13 mm and 10 mm spheres). Therefore, 20 subsets were applied in this study. Among all reconstruction parameters, the FWHM of the Gaussian filter showed the most significant correlations (all Pearson's r < − 0.85, all P values < 0.05) with RCs as well as the highest regression coefficients (− 9.49 to − 11.83 for the 32:1 T/B ratio, − 8.68 to − 11.83 for the 16:1 T/B ratio, − 6.23 to − 10.90 for the 8:1 T/B ratio, − 4.23 to − 9.39 for the 4:1 T/B ratio, respectively, Table 1). This indicated that the Gaussian filter had a large impact on quantifications. This study showed that there was no plateau of FWHM in terms of RCs, and RCs decreased dramatically along with the decrease of FWHM in all T/B ratios. Since there was no plateau for the Gaussian filter in terms of RCs, it was not used in the following analysis.
The AC + SC + RR combination presented a higher concentration concordance in spheres, and the AC + SC combination resulted in the lowest RCs ( Fig. 5 and Table 2). Although RCs calculated with the AC + RR combination were higher than those of For RCs of the largest 37 mm sphere calculated with the AC + RR combination in all T/B ratios, the plus tolerances were even greater than 20%. In contrast, these plus tolerances were only approximately 2.89-6.70% for the AC + SC + RR combination ( Table 2). Since the scattered photons account for 30-40% of all photons acquired by the SPECT detector, the application of SC can reduce the errors of the calculated concentration to a great extent [61].
In principle, and among other factors, image quality in nuclear medicine is mainly affected by three factors: (1) spatial resolution (image sharpness) [62], (2) noise (variations in the image due to random effects such as quantum noise) [63], and (3) contrast (difference in image intensity between areas of the imaged object) [64]. The per cent contrast only measures the contrast aspect of the images but does not measure noise or spatial resolution, which altogether affects the image quality, as measured by lesion detectability. The resolution in this study was unchanged since the phantom did not have much anatomical variability. Thus, we added a noise evaluation (COV analysis) to the study in addition to the per cent contrast analysis to make the study more comprehensive and the results more indicative or potentially applicable to clinical situations. For image quality, the combination of AC + SC + RR had the best per cent contrasts in all T/B ratios (Table 3, all P values < 0.05). However, the COVs of the AC + SC + RR combination were higher than those of the AC + RR combination at all T/B ratios (Table 4, all P values < 0.05). The study of Knoll et al. also showed a similar result that the application of SC might increase the background variability [64]. This indicated that although AC + SC + RR combination resulted in the best quantitative performance, its image quality might be somewhat debatable. Since quantification was the main aim of this study, we selected AC + SC + RR combination as the optimal correction combination. Several reports also suggested the significance of AC, SC, and RR for SPECT quantification [29,[65][66][67]. Our study also showed that for all T/B ratios, COVs were relatively high. This was an inevitable compromise when quantification was the main aim of this study since a larger number of iterations could not only increase quantification accuracy but also increase background noise [47,68]. For the 32:1 T/B ratio, RCs under the 15% energy window were higher than those under the 20% energy window, and there was a statistically significant difference (P value = 0.023). However, for lower T/B ratios (16:1, 8:1, and 4:1), there were no statistically significant differences (all P values > 0.05, Table 5). This suggested that although CZT SPECT/CT has a better image resolution due to the improved energy resolution of the new solid-state crystals [69], for RCs, the advantage of the 15% energy window might not be obvious enough for lower T/B ratios compared with that of the 20% energy window. Since a 15% energy window might improve the quantification accuracy for a higher T/B ratio (32:1), it was determined to be the optimal energy window in this step.
This study showed that the correlations between RCs and acquisition time were not obvious compared with those of other parameters. The regression coefficients between RCs and the acquisition time/frame were relatively small (− 0.01 to 0.07 for the 32:1 T/B ratio, − 0.04 to 0.28 for the 16:1 T/B ratio, − 0.31 to 0.12 for the 8:1 T/B ratio, and − 0.34 to 0.26 for the 4:1 T/B ratio, respectively, Table 1). RCs did not increase significantly with increasing acquisition time/frame (all P values > 0.05 when comparing 10 s with 120 s of acquisition time/frame). However, for the 4:1 T/B ratio, RCs of six spheres increased significantly within the first 40 s/frame of acquisition time (10 s vs. 120 s, P value = 0.015; 40 s vs. 120 s, P value = 0.060, Fig. 7). This suggested that for lower concentrations, the optimal acquisition time might be dependent on the activity concentration. Meanwhile, SD could be rapidly reduced within the first 40 s/frame of acquisition time (Fig. 8). These results suggested that acquisition time might only impose a strong impact on the quantification accuracy within this range. Therefore, 40 s/frame might be the optimal value in this step. In practice, 40 s/frame of acquisition time might be able to satisfy the quantification requirement.
The RCs of the 37 mm sphere reached 102.02-107.65% under the best acquisition and reconstruction parameters. However, these numbers dropped dramatically as the sphere volumes decreased (55.07-79.02% in the 10 mm sphere, Table 6). One possible reason is that the VOI counts decreased significantly with smaller objects due to the limitations brought on by the spatial resolution of SPECT [70].
There were four limitations to this study. First, the suggested acquisition and reconstruction parameters might only be applicable for quantitative purposes and for the CZT-SPECT equipment we investigated in this study. Second, due to the purpose of calculating mean uptake values with the lowest errors and determining the impacts of different acquisition and reconstruction parameters, VOIs were delineated using CT images as references. This procedure might be limited in clinical usage. Third, this study also showed that for all T/B ratios, COVs were relatively high. This was an inevitable compromise when quantification was the main aim of this study since a larger number of iterations could not only increase quantification accuracy but also increase background noise. Last, quantification measurements were only performed with a CZT-based camera system and not with a NaI (Tl)-based camera system, so a comparison between them was not evaluated.

Conclusions
CZT-SPECT/CT of technetium showed a good quantification accuracy. The favourable acquisition parameters may be the 15% energy window and 40 s/frame. The favourable reconstruction parameters could be 35 iterations, 20 subsets, the AC + SC + RR correction combination, and no filter. Our results might have some merit for clinical quantification guidelines.