Patients

This study included 56 patients (age, 71 ± 11 years; male, 79%) with definite or suspected CAD who were assessed by myocardial perfusion imaging (MPI) (Table 1). Old myocardial infarction was clinically diagnosed in 24 (44%) patients (body mass index [BMI], 23 ± 5 kg/m2) who were included for comparisons of perfusion at rest. The summed rest score (SRS) determined automatically using standard databases by the Japanese Society of Nuclear Medicine normal database was 6.9 ± 6.6 (range 0-24).7

Table 1 Demographics of patientsImaging protocol

The resting part of stress-rest MPI was used from a one-day protocol study. All patients fasted during the morning and refrained from taking any medications that might affect the MPI results. The patients were then assessed under stress by adenosine SPECT imaging using 260 MBq (7 mCi) of 99mTc-methoxy-isobutyl-isonitirile (MIBI) (n = 49) or tetrofosmin (n = 7), and at rest with 700 MBq (18 mCi) of the same tracers.

Acquisition of SPECT data

Conventional SPECT images were acquired at Kanazawa University Hospital using a Symbia Intevo 16 SPECT/CT and our routine clinical protocol comprising of AutoForm™ low-energy high resolution collimators with a 15% peak window, centered at 140.5 keV for 99mTc and a 15% lower abutting scatter window (standard dual-energy window method). Sixty projection views over 360° with 6° step-and-shoot sampling were accumulated using a 64 × 64-matrix with a zoom factor of 1.45 (6.6-mm pixels), 35 seconds per view in a tight circular orbit (rotation radius of 24 cm).

A 3%-NIST traceable calibrated 57Co-source was used to calibrate the SPECT imager per xSPECT Quant TM protocol. The prototype Footnote 1 xSPECT Cardiac (xSC) acquired in continuous rotation and list-mode with 120 projection views over 360° with 3° sampling in a 256 × 256 matrix (2.4-mm pixels), 15 seconds per view, non-circular orbit. Thus, the double-scan procedure consisted of current clinical and research protocols. The sequence of scan was conventional resting SPECT followed by xSC data acquisition. The R-wave served as a trigger for conventional prospective electrocardiogram (ECG) gating with a fixed width beat window (20%) and enabled auto-tracking of RR intervals that are divided into 16 gated frames. Nuclear medicine technologists appropriately set a wider acceptance window for patients with arrhythmia such as frequent premature beats and atrial fibrillation. In the xSC list-mode acquisition, the ECG measured physiological trigger events allowed gated continuous rotation followed by retrospective gating during reconstruction.

SPECT image reconstruction

We used our clinical standard Flash3DTM for conventional ungated and gated reconstructions, with a distance-dependent, 3D-Gaussian collimator model for the point spread function within OSEM at 120 updates (12 subsets and 10 iterations) based on corrected framed data. A 3D-Gaussian filter at a full-width at half maximum (FWHM) of 13.2 mm was used for post smoothing. We do not typically use CT attenuation correction in clinical practice at our hospital.

The xSC reconstruction was based on xSPECT QuantTM from list-mode data using a measured 3D point response function, 3D rotation with adaptive class-standard gantry deflection, CT-derived attenuation correction (AC), and energy window-based scatter correction (SC) as described by Vija et al..8,9,10 In this work, we chose the OSEM-based xSPECT Quant reconstruction, as it yields a more similar resolution and noise impression to the Flash3D reconstruction.

The xSC reconstruction contained retrospective gating, inter-viewFootnote 2 and intra-viewFootnote 3 motion correction, which are detailed below.

Retrospective gating: Retrospective gating enabled efficiency improvement by allowing continuous rotation in gated studies thereby maximizing the counts utilized for each view. In the xSC reconstruction, a beat histogram was curated using automated data analysis for outliers,11 where the acceptance window was defined with ± 3 root-mean square of the median in an iterative process run to convergence (“curation”). The curation rejected so defined premature ventricular contraction and 1 beat after it. This process was an attempt to maintain the maximum number of counts possible even when abnormal cardiac rhythms presented a challenge to the prospective approach. In addition, it reduces tomographically inconsistent assignments of gate to angle that may occur in the prospective scheme resulting in, e.g., radial blurring and then potentially impacting LVEF assessment.12 Albeit any number gates are in principle possible, we also selected 16 gated frames as a good trade-off between volume curve sampling and statistical power. Each xSC gate was reconstructed with attenuation and scatter correction.

We checked for beat histogram consistency in the back-to-back clinical and dedicated-research acquisitions, to exclude a temporal bias due to sequenced acquisition. We applied the same prospective gating and framing algorithm to the list-mode data as implemented in the clinical product which framed them with a 20% acceptance window and auto-tracking. We examined six patients with frequent arrhythmia and outlier beat lengths, and 12 without arrhythmia during data acquisition.

Inter-view motion “correction”: Its goal was to reduce tomographic inconsistencies, mitigating a build-up of image artifacts as updates progress. It was designed to be automated and compatible with iterative reconstruction. However, it is limited by poor count statistics and the fundamental limitations of the image formation. The key difference between conventional and this new approach is briefly described in the following (see details in supplement file Section A). The current commercially available proprietary semi-automatic method was based on estimation of axial and transaxial motion in each projection view to mitigate cardiac motion induced by patient motion or cardiac creep. Transaxial and vertical motion estimations were based on sinograms and linograms, respectively, and the projections were shifted view-by-view in axial and transaxial direction prior to reconstruction.13,14 Fundamentally, the method suffered when significant change of contrast of the myocardial projection occurred due to patient-specific factors. The contrast can fall so low that edge detection and subsequent fits become unreliable in some views, and potentially inducing erroneous shifts in these views. Manual intervention is often needed, which is subjective and rarely repeatable, resulting in a trial-and-error reconstruction approach. The inter-view motion correction overcomes this issue using a fully automated method applicable to all study types including cardiac studies, which allows for rigid axial and transaxial shifts measured within and was designed for the iterative reconstruction process.15 The shift vectors were estimated in a dedicated automated reconstruction preceding the user-adjustable target reconstruction. This dedicated reconstruction named “Improving Tomographic Consistency” (ITC)-reconstruction used the same update method, data model, and enables AC and SC as the target xSC reconstruction. It was not user adjustable and solely for the purpose of shift vector estimation. The xSC reconstruction restarted using the shift vectors from the preceding ITC reconstruction by updating the 3D rotation matrix, thereby improving tomographic data model consistency for all views. Strictly speaking this approach represents a motion correction only if the tomographic inconsistency was caused by a rigid-body axial and transaxial motion, otherwise it is merely a mitigation.

Intra-view motion “correction”: This refers to mitigating for motion effects during the acquisition of a view for the duration of the dwell time at the specific viewing angle (see details in supplement file, Section B). The method can mitigate the impact of motion during the dwell time for both continuous and discontinuous rotation if the statistical data quality allows it. Respiration is typically the main cause for such motion. However, respiratory-induced cardiac motion may also show up as view-to-view jitter, a type of inconsistency appearing over neighboring views. This type of residual inter-view jitter is more noticeable when the dwell time is on the order of the respiration period, which can be addressed by the subsequent use of inter-view motion correction. We used the non-linear Laplacian Eigenmap data-driven dimensionality reduction approach based on the assumption that the respiration information data laid on a lower-dimensional manifold.16,17,18 The data-driven gating subdivided the projection view into respiratory-gated views (RGV). In essence, the dwell time was split into six estimated respiratory gates of unequal temporal duration to create six RGV’s, enabling an estimate of the center-of-light Footnote 4 (COL) motion of the field of view (FOV) in each RGV, which were then shifted to a common reference location within the FOV, prior to summation of all RGV’s into the projection frame that was subsequently used in the reconstruction. This approach attempted to mitigate respiration-induced blurring and shape deformation, and was limited by poor count statistics and, thus, mandatory limitations of the number of degrees of freedom in the underlying model. Thus, the method only allowed for axial rigid translation.

Effect of motions on cardiac phantom

We assessed the potential improvement on image quality of the inter and intra-view motion correction method by applying patient-derived shift vectors to a perfectly still phantom dataset, thereby inducing the artifacts expected of the respective type of inconsistency. We used data from an anthropomorphic normal male cardiac phantom (Data Spectrum Corporation, Durham, NC, USA) acquired in the same configuration as the xSC. The shift vectors were extracted from patients, and histograms of all the transverse (x) and axial (y) shift vectors of the patient data were generated.

Visual analysis of image quality

Two experienced nuclear medicine physicians and a technologist visually assessed vertical and horizontal long- and short-axis images reconstructed by conventional and xSC methods. They compared the quality of blinded images, noise, artifacts, the possibility of false defects and a previous myocardial infarction if it had been clinically diagnosed prior to this study.

Quantitative analysis

All beat lengths during SPECT data acquisition were analyzed using RR interval histograms, and fractions of beats within the 20%-acceptance window beat length were measured. Images were reconstructed and displayed as long- and short-axis images using 4DM software (INVIA, Ann Arbor, MI, USA) with standard 17-segment polar map display and segmental count (%) calculations. Left ventricular (LV) end-diastolic volume (EDV), end-systolic volume (ESV), and ejection fraction (EF) were compared between prospective and retrospective gating methods. Echocardiographic measurements of volume and EF (modified Simpson method with four- and two-chamber views) within 3 months of an MPI study served as the reference. Cross-sectional and circumferential profiles were analyzed at the mid-portion of transaxial SPECT images as well as slices with perfusion defects to evaluate contrast.

Ethics Committee approval

The Ethics Committee of Kanazawa University approved this collaborative study. All included patients provided written informed consent to participate.

Statistics

All data are expressed as means ± standard deviation (SD), and means were compared using the analysis of variance. Categorical variables were analyzed using contingency tables. Segmental counts on polar maps were assessed using paired-comparison tests. Linear regression was calculated using the least-squares method. All data were statistically analyzed using JMP Pro v. 16.0 (SAS Institute Inc., Cary, NC, USA). Values with P < 0.05 were considered significant.