Impact of iterative reconstruction with resolution recovery in myocardial perfusion SPECT: phantom and clinical studies

The corrections of photon attenuation, scatter, and depth-dependent blurring improve image quality in myocardial perfusion single-photon emission computed tomography (SPECT) imaging; however, the combined corrections induce artifacts. Here, we present the single correction method of depth-dependent blurring and its impact for myocardial perfusion distribution in phantom and clinical studies. The phantom and clinical patient images were acquired with two conditions: circular and noncircular orbits of gamma cameras yielded constant and variable depth-dependent blurring, respectively. An iterative reconstruction with the correction method of depth-dependent was used to reconstruct the phantom and clinical patient images. We found that the single correction method improved the robustness of phantom images whether the images contained constant or variable depth-dependent blurring. The myocardial perfusion databases generated from 72 normal patients exhibited uniform perfusion distribution of whole myocardium. In summary, the single correction method of depth-dependent blurring with iterative reconstruction is helpful for myocardial perfusion SPECT.

perfusion distribution was not determined when the number of iterations was change. Our motivation for this research was to provide the uniform count distribution with less depth-dependent blurring in myocardial perfusion SPECT images.
The aim of this study was to confirm that the myocardial perfusion SPECT distribution derived from OSEM RR in both phantom and clinical studies reduces the differences between circular and non-circular orbits compared to reconstruction with FBP. Consequently, we generated a normal database (NDB) for OSEM RR derived from patients with a low likelihood of coronary artery disease (CAD) and evaluated myocardial perfusion distribution.

Materials and Methods
cone-shape phantom. An acrylic cone-shape phantom has a height of 112 mm and a basal diameter of 66 mm (Fig. 1a). The compartment for the radioisotope was filled with 64 MBq of 99m Tc-pertechnetate, and an inner cavity was filled with non-radioactive water. The volumes of compartments for radioisotope and water were 118 mL and 93 mL, respectively. The noncircular orbit with the 360-degree rotation of gamma cameras was applied to the phantom acquisition in close proximity to the surface of the phantom and bed (Fig. 1b). A circular orbit was also applied in this phantom study.
Anthropomorphic torso phantom. We utilized the anthropomorphic torso phantom configured with the cardiac, pulmonary and hepatic components (Kyoto Kagaku, Kyoto, Japan). The left ventricular (LV) myocardium and liver were filled with 185 and 19 MBq of Tc-99m pertechnetate, respectively. The left and right ventricular cavities were filled with non-radioactive water. The circular-orbit acquisition (CA) and noncircular-orbit acquisition (NCA) with the 360-degree rotation of gamma cameras were used for the anthropomorphic torso phantom study. Four plastic circular defects with a 20 mm diameter were placed in the mid-anterior, lateral, inferior, and septal walls. Study population. We retrospectively included 37 female and 35 male patients with a low likelihood of CAD from Kanazawa University Hospital. 99m Tc-sestamibi or 99m Tc-tetrofosmin MPI was performed with NCA of gamma cameras. The mean age of male and female patients was 68 ± 16 and 67 ± 15 years, respectively (p = n. s.). The mean height and weight were 151 ± 5 cm and 50 ± 6 kg for females, and 165 ± 8 cm and 66 ± 15 kg for males, respectively (p < 0.0001 for both). Approval for this study was obtained from the ethics committee of the Kanazawa University. The current study was performed in accordance with the ethical guidelines of the Declaration of Helsinki. Due to our retrospective study, informed consent was obtained from all subjects in the form of opt-out. Patients who rejected participation in our study were excluded. image acquisition and data processing. In both phantoms and clinical studies, SPECT acquisition was performed with a dual-head gamma camera (Symbia T6 hybrid SPECT/CT scanner, Siemens Healthcare K.K., Tokyo, Japan) equipped with a low-energy high-resolution collimator. The pixel size was 6.6 mm for a 64 × 64 matrix. A photopeak energy window of 99m Tc was set at 15% centered at 140 keV. In the phantom study, acquisition time was set as 5 seconds per projection. 60 projection images with CA and NCA of gamma cameras were obtained. In the clinical study, we performed stress and rest gated MPI with 16 frames per cardiac cycle on the hybrid SPECT/CT scanner. MPI was performed with a 360-degree NCA with 60 projections at 40 minutes after injection of 99m Tc tracers of 300-370 MBq. The acquisition time was set as 35 seconds per projection. Data analysis. We used two-reconstruction processing methods: OSEM RR and filtered back projection (FBP) in the phantom and clinical studies. The resolution recovery processing is based on a Flash 3D algorithm (Siemens Healthcare K.K., Tokyo, Japan) 29 . The Flash 3D algorithm was not incorporated into the FBP processing. In the phantom study, when the subset was constantly set as 15, the number of iterations was set at 1, 5, 10, 15, and 20. In the clinical study, subsets and iterations were set as 15 and 8, respectively. We utilized a Gaussian post-filter for OSEM RR and a Butterworth filter for FBP in both studies. The full width at half maximum of the Gaussian filter was 13.2 mm. The cutoff frequency and order for the Butterworth filter were 0.68 Nyquist and 8. All OSEM RR and FBP processing was performed using e.soft version 8.1 (Siemens Healthcare K.K., Tokyo, Japan).
In the cone-shape phantom and anthropomorphic torso phantom studies, circumferential profile analysis was used to assess SPECT count profiles 30,31 . The SPECT count profile was presented on a line chart as a function of angle, the origin of which corresponds to the 12 o' clock position on the phantom images. The angles were measured in a clockwise direction relative to an origin at 12 o' clock. The anterior, lateral, inferior, and septal regions of myocardium were defined as 315-45°, 45-135°, 135-225°, and 225-315° in the anthropomorphic torso phantom, respectively. The defect regions of myocardium were also defined as 345-15° for the anterior wall, 75-105° for the lateral wall, 165-195° for the inferior wall, and 255-285° for the septal wall in the anthropomorphic torso phantom with defect, respectively. The SPECT count profiles were evaluated in the central transaxial slice of the cone phantom and the central short axial slice of the anthropomorphic torso phantom. A count ratio of anterolateral to the anteroseptal region was calculated as segments 8 and 14 divided by segments 12 and 16 in the quantitative analysis of the NDB using a standard 17-segment model 32 . The count ratio of the anterolateral to the inferior region and that of the anterolateral to the anterior region were also calculated as segments 10 and 15 divided by segments 12 and 16, and segments 7 and 13 divided by segments 12 and 16, respectively. Polar maps were generated with quantitative perfusion SPECT (QPS) version 2008.1 (Cedars-Sinai Medical Center, Los Angeles, CA, USA).
The compatibility of Gaussian and Butterworth filters was evaluated in the anthropomorphic phantom study. When the cutoff frequency of Butterworth filter was set as 0.68 Nyquist in our clinical condition, the FWHM of Gaussian filter was changed from 3.3 to 16.5 mm. SPECT count profiles for the central short axial slice of the cardiac component were compared in both Gaussian and Butterworth filters. The comparative evaluation of OSEM RR and FBP reconstructions with or without Gaussian filtering was also performed. SPECT count distributions were calculated for OSEM RR and FBP images without Gaussian filtering, and a fast Fourier transform was used to convert the SPECT count distributions into the frequency domain using ImageJ version 1.52p (NIH, Bethesda, Maryland, USA).

Statistical analyses.
All continuous values were expressed as a mean ± standard deviation (SD). The differences in continuous variables were analyzed using Student's t-test and Tukey-Kramer test. A paired t-test was www.nature.com/scientificreports www.nature.com/scientificreports/ used to analyze the differences in paired continuous data. All statistical tests were two-tailed, and a p-value of less than 0.05 was considered significant. These analyses were performed using the JMP version 11.2.1 (SAS Institute Inc., Cary, NC, USA) software.

Results
cone-shape phantom. Figure 1 shows a cone-shape phantom and experimental condition. The projection images were obtained from NCA (Fig. 1b) and CA of gamma cameras. The rotation radius was constantly 230 mm in CA, and minimum and maximum radii were 91 mm and 230 mm in NCA, respectively. The superior resolution of projection images was visually shown at 0 and 180 degrees of gamma camera in NCA compared with CA (Fig. 1c). Acquired counts of the projection data ranged from 111 kilocounts/view at 91 mm to 62 kilocounts/ view at 230 mm in NCA. The projection data acquired with circular and noncircular orbits of gamma cameras were reconstructed with FBP and OSEM RR (Fig. 2). Polar maps derived from CA and NCA were different greatly in FBP (Fig. 2a,c). Polar maps derived from CA and NCA were visually similar in OSEM RR with 10, 15, and 20 iterations except for the apex. Normalized SPECT count at the front region was higher in NCA than in CA in Supplementary Fig S1. An inner space of the cone phantom derived FBP and OSEM RR with one iteration was smaller in CA than in NCA (Fig. 2b,d). That derived from OSEM RR with 20 iterations was equivalent in CA and NCA. OSEM RR with 5, 10, 15, and 20 iterations yielded similar normalized count profiles in CA, and that with 10, 15, and 20 iterations also yielded similar normalized count profiles in NCA (Fig. 2e,f). The largest and smallest normalized count differences were exhibited in FBP (0.16 ± 0.11) and OSEM RR with 20 iterations (0.0026 ± 0.0059), respectively (p < 0.0001, Fig. 2g,h).
Anthropomorphic torso phantom: normal myocardium model. The polar maps of the anthropomorphic torso phantom were generated to confirm the difference between FBP and OSEM RR in both CA and NCA. Polar maps derived from CA and NCA were visually similar in FBP and OSEM RR (Fig. 3a,b). OSEM RR with 10, 15, and 20 iterations yielded similar normalized count profiles in CA and NCA (Fig. 3c,d). A lower normalized count difference was exhibited in OSEM RR with 10 iterations (−0.0061 ± 0.029), whereas higher normalized count differences were exhibited in FBP (0.039 ± 0.058) and OSEM RR with 1 iteration (0.040 ± 0.037), www.nature.com/scientificreports www.nature.com/scientificreports/ (p < 0.0001, Fig. 3e,f). No significant differences between CA and NCA were exhibited in FBP and OSEM RR with 1 and 5 iterations in the anterior wall (Fig. 3g), in FBP and OSEM RR with 10, 15, and 20 iterations in the lateral wall (Fig. 3h), in OSEM RR with 10, 15, and 20 iterations in the inferior wall (Fig. 3i), and in OSEM RR with 1, 5,10, 15, and 20 iterations in the septal wall (Fig. 3j).
Anthropomorphic torso phantom: myocardial defects model. The polar maps of the anthropomorphic torso phantom with myocardial defects were generated to confirm the difference between FBP and OSEM RR in both CA and NCA (Fig. 4a,b). FBP and OSEM RR with 5, 10, 15, and 20 iterations yielded similar normalized count profiles in CA and NCA (Fig. 4c,d). Significant higher normalized count differences were exhibited in FBP (0.044 ± 0.038) and OSEM RR with 1 iteration (0.043 ± 0.029) than in OSEM RR with 20 iterations (0.009 ± 0.030) (p < 0.0001 for all, Fig. 4e,f). No significant differences between CA and NCA were exhibited in OSEM RR with 5, 10, 15, and 20 iterations in the anterior (Fig. 4g), lateral (Fig. 4h), inferior (Fig. 4i), and septal walls (Fig. 4j). To confirm the compatibility of Gaussian and Butterworth filters, the phantom delineation created from Gaussian filter with 3.3-, 6.6-, 8.3-, 9.9-, 11.6-, 13.2-, 14.9-, and 16.5-mm FWHMs was compared with that from 0.64 Nyquist frequency of Butterworth filter in Supplementary Fig. S2. The smallest count difference was exhibited in the combination of Butterworth with 0.64 Nyquist frequency and Gaussian with 13.2-mm FHWM. OSEM RR and FBP reconstructions with or without Gaussian filtering was also compared in Supplementary Fig. S3. The amount of statistical noise component without Gaussian filter was higher in FPB than in OSEM RR . Although the effect of Gaussian filter for the noise reduction was similar in FBP and OSEM RR images, the SPECT count distribution for FBP significantly differed from that for OSEM RR in the mid and basal regions. clinical subjects. The NDBs were generated to confirm the difference of clinical myocardial perfusion distribution between FBP and OSEM RR in NCA (Fig. 5a,b). Myocardial perfusion counts were always high in the anterolateral region in both male and female NDBs. Moreover, those showed higher values in the anteroseptal region in OSEM RR . The segmental myocardial perfusion distributions derived from OSEM RR paralleled those from FBP in both male and female NDBs (Fig. 5c,e). Based on count ratios of anterolateral segments to anteroseptal www.nature.com/scientificreports www.nature.com/scientificreports/ segments, inferior segments, and anterior segments (Fig. 5d,f), significantly higher myocardial perfusion counts were exhibited in OSEM RR than in FBP. The clinical study demonstrated that slight ischemia was delineated in the inferior regions in FBP and OSEM RR with 10, 15, and 20 iterations (Fig. 6a). The delineation of the LV cavity by OSEM RR with 20 iterations was superior to that by FBP (Fig. 6b).

Discussion
This present study showed that OSEM RR could yield similar count distributions in CA and NCA in the phantom study. In the anthropomorphic torso phantom study with defects, defect delineation was equivalent between CA and NCA in OSEM RR processing. Clinical myocardial perfusion showed a more uniform distribution in OSEM RR than FBP in the NDB. Since the rotation radius varies greatly in SPECT acquisition, we would suggest the use of OSEM RR in nuclear cardiology study. However, the OSEM RR processing with 10, 15, and 20 iterations should need to yield superior myocardial perfusion.
Although our study revealed that OSEM RR compensated for myocardial perfusion counts of NDBs in the anterior, anteroseptal, and inferior walls, uniform myocardial count distribution in the whole myocardium has not yet been shown in OSEM RR processing. Even if attenuation and scatter corrections are used in MPI, uniform myocardial perfusion would not be observed 17,33 because the anterior and inferior myocardial counts are influenced by respiratory motion owing to diaphragmatic movement in digital phantom studies [34][35][36] . Regarding the myocardial count distribution, maximum myocardium movement is observed in the apex 37 , and then myocardial 99m Tc distribution will be blurred. Consequently, further corrections for cardiac and respiratory movements are needed to find the true myocardial perfusion distribution.
The OSEM technology can easily be integrated with the RR algorithm. Furthermore, our results showed no disadvantages and artifacts owing to using OSEM RR in the phantom scans and clinical NDBs. Although the FBP has commonly been used to reconstruct slices from projection images in nuclear cardiology, we highly www.nature.com/scientificreports www.nature.com/scientificreports/ recommend the use of OSEM RR in daily clinical practice. Moreover, manufacturers have developed their own version of RR algorithms for OSEM iterative reconstruction, such as Astonish by Philips Medical Systems, Milpitas, CA, USA 38 , Evolution for Cardiac by GE Healthcare, Waukesha, WI, USA 39 , and wide beam reconstruction by UltraSPECT, Haifa, Israel 3,39 . Consequently, the optimal number of iterations and filtering parameters should need to be investigated by OSEM RR fitted for each SPECT imaging system.
The effects of OSEM RR in CA and NCA were evaluated in the cone-shape phantom, anthropomorphic torso phantom, and clinical patient studies. Although the orbit of the camera for the cone phantom experiment did not represent a clinical condition, our aim of this experiment was to confirm the SPECT count distribution of RR in SPECT acquisition using a simplified model. In the anthropomorphic torso phantom without defects, count distributions between FBP and OSEM seemed to be similar. However, defects delineation was different between CA and NCA in FBP. Myocardial perfusion counts of clinical NDB were corrected with OSEM RR in the deep regions of the heart: anterior, anteroseptal, and inferior walls. The OSEM RR processing could contribute to producing uniform myocardial perfusion counts even without the attenuation correction.
Our study has several limitations. We did not compare CA and NCA in the clinical study. No optimal reconstruction condition for the OSEM RR was used in the clinical studies. However, a small number of OSEM iterations yielded insufficient compensation for the RR in the phantom and clinical studies (Figs. 2-4 and 6). Moreover, optimal smoothing filtering conditions for Gaussian and Butterworth filter was not evaluated. This is because OSEM RR incorporating Butterworth filter is not supported by the Siemens data processing system. However, the effect of smoothing filtering condition, which was used in this study, has already been examined in our previous study 16 . Since quantitative scoring of myocardial perfusion using NDBs was not performed in the present clinical study, summed stress/rest/difference scores could be calculated to confirm the difference between OSEM RR and FBP.

conclusion
We clarified the myocardial perfusion distribution corrected with RR in CA and NCA of SPECT imaging in the cone-shape and anthropomorphic torso phantom studies. The phantom study characterized the relationship between iterations and count distribution in OSEM RR processing. Moreover, the count distribution derived from NCA paralleled those from CA. When NDBs were created to evaluate the myocardial perfusion distributions in clinical studies, OSEM RR is recommended to compensate for the depth-dependent blurring in myocardial perfusion counts on the 17-segment polar map. Figure 6. Clinical stress-rest polar maps (a) and SPECT images (b) in a 60-year-old male patient with slight ischemia in the inferior wall. Height and weight were 167 cm and 66 kg, respectively. All images were acquired with the noncircular orbit of gamma cameras. The Iterations changed from 1 to 20, and a subset was constantly 15.