Triple-gated motion and blood pool clearance corrections improve reproducibility of coronary 18F-NaF PET

To improve the test–retest reproducibility of coronary plaque 18F-sodium fluoride (18F-NaF) positron emission tomography (PET) uptake measurements. We recruited 20 patients with coronary artery disease who underwent repeated hybrid PET/CT angiography (CTA) imaging within 3 weeks. All patients had 30-min PET acquisition and CTA during a single imaging session. Five PET image-sets with progressive motion correction were reconstructed: (i) a static dataset (no-MC), (ii) end-diastolic PET (standard), (iii) cardiac motion corrected (MC), (iv) combined cardiac and gross patient motion corrected (2 × MC) and, (v) cardiorespiratory and gross patient motion corrected (3 × MC). In addition to motion correction, all datasets were corrected for variations in the background activities which are introduced by variations in the injection-to-scan delays (background blood pool clearance correction, BC). Test–retest reproducibility of PET target-to-background ratio (TBR) was assessed by Bland–Altman analysis and coefficient of reproducibility. A total of 47 unique coronary lesions were identified on CTA. Motion correction in combination with BC improved the PET TBR test–retest reproducibility for all lesions (coefficient of reproducibility: standard = 0.437, no-MC = 0.345 (27% improvement), standard + BC = 0.365 (20% improvement), no-MC + BC = 0.341 (27% improvement), MC + BC = 0.288 (52% improvement), 2 × MC + BC = 0.278 (57% improvement) and 3 × C + BC = 0.254 (72% improvement), all p < 0.001). Importantly, in a sub-analysis of 18F-NaF-avid lesions with gross patient motion > 10 mm following corrections, reproducibility was improved by 133% (coefficient of reproducibility: standard = 0.745, 3 × MC = 0.320). Joint corrections for cardiac, respiratory, and gross patient motion in combination with background blood pool corrections markedly improve test–retest reproducibility of coronary 18F-NaF PET.


INTRODUCTION
Positron Emission Tomography (PET) in combination with Computed Tomography (CT) angiography (CTA) holds promise as a non-invasive technology for identification of high-risk plaques in patients with coronary artery disease [1][2][3][4].
Clinical implementation of coronary PET imaging is, however, challenging, as coronary lesions are small and continuously move during the acquisition.Further, only modest target to background ratio (TBR) differences between culprit and non-culprit plaques (~34%) have been reported for 18 F-sodium fluoride PET ( 18 F-NaF) [1].
Importantly, the TBR measurements are significantly degraded by cardiorespiratory and patient motion during the 30-min scans.It has been shown that physiological tidal breathing can cause the heart to move >1 cm [5].The amplitude of coronary artery motion during the cardiac cycle is about 8-26 mm, depending on the artery and location, with the highest motion in the right coronary artery [6].Further, typical gross patient motion (other than cardiorespiratory motion) results in repositioning of the heart, typically by 5-15mm, during a 30-min scan [7].These observations are of key significance for coronary lesions with dimensions measured in single millimeters.
To reduce the effect of motion of the coronary arteries, end-diastolic phase images have been selected in studies to date [1,6,8] but this strategy uses only 25% of PET counts, consequently increasing image noise [9].Recent studies proposed improvements by correcting for cardiac motion [6,9], or by combining end-diastolic imaging with corrections for gross patient motion [7], but did not include corrections for respiratory motion, nor test how these corrections affect the scan-rescan reproducibility.Additionally, TBR values are affected by variations in the tracer uptake time (injection to scan delay) [10,11].
Consequently, the reproducibility of this promising PET technique remains suboptimal, which hampers its translation into clinical routine.
In this study, we demonstrate that a novel technique for coronary PET processing which combines triple motion correction (3xMC) (cardiorespiratory and gross patient motion) with adjustments for injection-to-scan delays significantly improves the scan-rescan reproducibility.

Study Population
The study population comprised of twenty patients who underwent repeated hybrid 18 F-NaF PET/CT examinations of the coronary arteries.Scans were repeated within 3 weeks as a part of the ongoing DIAMOND (Dual Antiplatelet Therapy to Reduce Myocardial Injury, NCT02110303) study.One patient was excluded from the study due to an incomplete saving of the list mode PET file (PET raw data).Patient characteristics are described in Table 1.
Inclusion criteria for the study included angiographically confirmed multivessel coronary artery disease defined as either previous revascularization or stenosis > 50%.
Exclusion criteria included: an acute coronary syndrome within 12 months prior to the examination, renal dysfunction (estimated glomerular filtration rate ≤30 mL/min/1.73m 2 ) and contraindication to CT-contrast media.This study was approved by the local investigational review board (Edinburgh, UK) and written informed consent was obtained from all subjects.(400 mg/mL) was administered to the patients with a flow of 5-6 mL/s after determining the appropriate trigger delay defined by a test bolus of 20 mL of contrast material.CTA studies were assessed visually for percent stenosis according to SCCT guidelines [12].

Image reconstruction
Five different PET image reconstructions were evaluated in this study: (i) a static reconstruction using all the acquired data (no-MC), (ii) end-diastolic reconstruction using 25% of the acquired data (standard) [1], (iii) cardiac motion corrected reconstruction (MC), (iv) combined cardiac and gross patient motion corrected reconstruction (2xMC), and (v) combined cardiac, respiratory and gross patient motion corrected reconstruction (3xMC).All reconstructions except the end-diastolic reconstruction were using 100% of the acquired data.
The five datasets were reconstructed with vendor provided software (JS-Recon12, Siemens, Knoxville, USA) from the PET list mode data (raw PET data).All PET image reconstructions were performed with corrections for time-of-flight and point-spread function, using 2 iterations, 21 subsets.The no-MC reconstruction was performed without any gating, whereas all other reconstructions were performed with 4 cardiac gates (time/phase based) (all datasets); the number of gross patient motion frames depended on the motion of the individual patients and scans (range 2-10 frames [7]) (2xMC, 3xMC), while the number of respiratory gates was fixed to 4 (amplitude-based gating [13]) (3xMC).Because the gross patient and respiratory motion was detected directly in sinogram space, it was not possible to apply any direct motion correction of the corresponding AC maps owing to the complex translations from the motion vectors obtained in projection space to image-space [7].

Motion detection
The overall scheme for motion detection is shown in Figure 1.
Cardiac gating information was obtained from a 3-lead electrocardiogram.Respiratory and gross patient motion detection was achieved using only the acquired PET list data without the use of any external markers.The data-driven motion detection techniques were based on center-of-mass analyses of single-slice rebinned sinograms [14] created for every 200 ms of the acquisition, as described in our previous study [7].In brief, the detection of the gross patient motion was obtained for the entire field-of-view [7], while the respiratory motion was detected only for the diaphragm using a 2cm (radius) boundary.

Motion correction
The PET motion correction was obtained through co-registration of gated PET-images (PET-PET image co-registration), using a dedicated coronary PET/CT software (FusionQuant, Cedars-Sinai Medical center) which employs a non-linear co-registration of the images [8] (Figure 1).The resulting motion corrected datasets (MC, 2xMC, and 3xMC) were obtained through averaging of the co-registered images.To ensure accurate co-registration of the coronary plaques, the motion compensation was focused on the coronary tree utilizing segmentations of the coronary arteries (including a radius of 1cm surrounding the center of the coronary arteries) obtained from the CTA images using a CT analysis tool (Autoplaque, Version 2.0, Cedars-Sinai Medical Center).

Image analysis
PET Quantification.Prior to image analysis, PET and CTA reconstructions were reoriented, fused and co-registered in all 3 planes (a rigid X-Y-Z translation was performed).Key points of reference were the sternum, vertebrae, blood pool in the left and right ventricle (based upon high 18 F-fluoride activity in the blood pool in comparison to the surrounding myocardium) and the great vessels [15]. 18F-NaF PET uptake was measured in all coronary segments with a CTA >25% stenosis, a vessel diameter ≥2 mm, which have not been stented and presented with image quality suitable for visual stenosis assessment.The 18 F-NaF uptake in the lesions was evaluated in the 3D spherical volume of interest (VOI) (radius 5 mm), and the background blood pool activity was determined by a cylindrical VOI (radius=10 mm, length=15 mm), placed in the right atrium at the level of the takeoff of the right coronary artery.We used the same VOIs for all 5 reconstructions evaluated in this study.TBR values for all 5 reconstructions were calculated by dividing the maximal standard uptake value (SUVmax) of the lesions by the mean SUV obtained from the blood pool (SUVBackground) [16].
The impact of the motion (cardiorespiratory and gross patient motion) was evaluated in three subsets of lesions, (a) in all CTA-defined lesion locations (b) in all lesions with 18 F-NaF-avid uptake and (c) in 18 F-NaF-avid lesions with gross patient motion >10 mm.The magnitude of the gross patient motion repositioning events was calculated in 3D from motion vectors obtained during the PET-PET co-registrations.
Blood pool correction.It has been recently reported that TBRs for 18 F-NaF varies with injection-to-scan time [10].Based on the data reported in [10], the decay-corrected tracer activity in the lesions does not change during a 2-hour period, while the blood pool activity is cleared at a rate of 1.5092*e -0.004*t (R 2 =0.81).From these findings, we propose to introduce a correction factor which harmonizes the SUVBackground activities to a reference time (60 minutes post-injection) (Equation 1).

Diagnostic evaluation of 18 F-NaF PET
All lesions were quantified based on the CTA based lesion position, categorized as 18 F-NaF-avid and 18 F-NaF-negative on standard PET using a previously validated methodology [1,17].In brief, lesions with TBR ≥1.25 and focal uptake on the site of the CTA-assessed lesion were considered 18 F-NaF-avid, while lesions without focal uptake or TBR values < 1.25 were considered 18 F-NaF-negative [1].

Statistical analysis
The data were tested for normality using the Shapiro-Wilk test.Statistical analysis was performed in MatLab (Mathworks).Continuous variables were presented as mean±SD (standard deviation).Assessment of the test-retest reproducibility before and after corrections for motion and blood-pool effects were obtained using descriptive statistics with Bland-Altman plots as well as the coefficient of reproducibility.
Improvements in the test-retest reproducibility were evaluated by Pitman-Morgan test.A p<0.05 was considered statistically significant.

Patient cohort
The patients underwent repeated 18 F-NaF PET / CTA hybrid imaging studies within a maximum of 21 days (mean 12±5 days).47 unique lesions were identified on the CTAimages with 15 18 F-NaF-avid, 30 18 F-NaF-negative and 2 lesions with discordant evaluations (TBR>1.25 in one scan, while TBR <1.25 in the other scan) on standard PET images.

Reproducibility with Motion Correction
The motion corrected datasets had significantly improved test-retest reproducibility of the lesion assessments in comparison to the standard datasets, all p<0.05 (Pitman-Morgan test) (Table 2).Figure 2 shows that the progressive motion compensation techniques steadily improved the test-retest reproducibility for all delineated lesions.

Background blood pool clearance correction
All datasets were corrected for the variances in the injection to scan delay by standardizing the tracer SUV measurements in the background region to 60-min after 18 F-NaF administration (Table 3).Due to the mean delay time (66±9 min, range 59-101 min) being longer than 60 minutes, our blood pool correction resulted in a slight reduction of TBR values by 2.5±3.8%(range: -0.4% to 17.8%), p=0.98 (Table 4).

Reproducibility with motion correction and BC
Blood pool correction further improved the test-retest reproducibility of TBR for all datasets.For all lesions as a stand-alone correction (Standard vs. Standard+BC), the coefficient of reproducibility was improved by 20%, in combination with 3xMC, the reproducibility was improved by 72% (Table 4, Figure 3).In the sub-analysis of 18 F-NaFavid lesions, 3xMC +BC correction improved the reproducibility by 78% (Table 4, Figure S1).Importantly in a subset of lesions with larger patient motion during one of the scans (>10 mm), 3xMC+BC correction lead to a 133% improvement in reproducibility (Table 5, Figure 4).
Two lesions with discordant assessment and two lesions considered 18 F-NaF-negative on the standard assessment were reclassified as being 18 F-NaF-avid on 2 scans following corrections for both cardiorespiratory and gross patient motion (3xMC).For the lesions with discordant analyses on test and retest scans (2 lesions), following 3xMC+BC corrections the TBR values increased from (TBR: standard = 1.24 and 1.22, on scan 1 and scan 2 respectively), to (TBR = 1.44 and 1.42, on scan 1 and scan 2 respectively).
Figure 5 displays two case-examples on the effect of the described motion correction techniques.In both cases, the detrimental motion caused discordant evaluations of the lesions in the test-retest evaluation.Following 3xMC and BC corrections, both lesions were reclassified with concordant evaluations ( 18 F-NaF-avid).

DISCUSSION
In this study, we evaluated the reproducibility of TBR measurements of coronary plaque activity before and after corrections for cardiorespiratory and gross patient motion as well as quantitative correction of the background activity according to the variations in the injection-to-scan delay.To our knowledge, this is the first time such comprehensive triple gating motion correction is reported for any PET imaging.We demonstrate that the reproducibility using the standard assessment (end-diastolic imaging) is impaired by motion during the acquisition, reduced count rate, and injection time variability.Motion corrected reconstructions utilizing all image data and adjustment for injection-to-scan delay markedly improved test-retest reproducibility.
High reproducibility of coronary PET lesion uptake is a critical prerequisite for the translation of coronary plaque imaging into clinical practice.The standard approach used to date is based on end-diastolic imaging [1,17].The rationale behind using end-diastolic images for the assessment of coronary plaque activity was to improve TBR and mitigate the detrimental effects of cardiac motion, as demonstrated in this study (Table 2) [1,18].
Unfortunately, this approach has several implications on TBR values, with two substantial problems pertaining to the increased noise in the images introduced by capturing counts from only one-fourth of the cardiac cycle and the remaining embedded motion-blur introduced by both respiratory and gross patient motion [7].These limitations result in compromised test-retest reproducibility, which consequently makes it difficult to set accurate TBR thresholds for positive or negative findings.In addition, the use of TBR as a clinical measure for the lesion uptake might not be ideal because of the varying uptake in the background introduced by variations in the injection-to-scan delays [10,11,19].In this study, the variations in the injection-to-scan delays were found to reduce the repeatability by 19% for all lesions (Table 3), which were corrected using a simple correction for blood clearance (BC).
To ameliorate the shortcomings of the noise and patient motion, motion correction of cardiac contraction has been employed [6,9,10].Additionally, in a recent study, it was found that the long acquisition duration (30 minutes) lead to multiple events of patient repositioning during emission scanning [7], a pattern that was also found in the current study.In this study, we combined corrections for gross patient motion with additional novel corrections for respiratory motion detected PET from list mode data to achieve total triple gated motion corrected reconstructions.These additional corrections improved test/retest reproducibility by an additional 12% to 42% in comparison to cardiac contraction correction (MC) as a standalone technique (Table 2).Overall, when BC and 3xMC were combined, the reproducibility was improved by 77% for 18 F-NaF-avid lesions (Table 4, Figures 3-4).
Despite significant improvement, the reproducibility remains modest (Coefficient of reproducibility of 25%) after applying the triple corrections proposed in this study.
Therefore, further reductions in the inter-scan variation is still warranted.The reproducibility coefficients were lower for the 18 F-NaF-avid lesions in the current study.This is expected to partly be caused by the background activities and partly by the increased noise in the SUVmax uptake for the 18 F-NaF avid lesions, a consequence of the quasi-Poisson nature of the PET-detections and single voxel of activity sampled [20].
While the use of SUVmean could reduce the noise, and thereby some of the variation in the measurements, it is not feasible as accurate delineation of small coronary plaques is impossible [19].An alternative method to reduce the noise in the TBR assessment could be the use of an average of the N hottest voxels within the lesion (SUVmax-N , with N>1), a method that has proven to be stable for oncological PET scans [21] in combination with more advanced motion correction techniques.However, the use of SUVmax-N mandates finding an optimum number of voxels and, thus, a new TBR cut-off value to determine the 18 F-NaF-avid lesions.This was outside of the scope of this study but should be considered for upcoming studies.
The combination of 3xMC and BC not only improved the reproducibility but also led to reclassification of 4 lesions, of which 2 were perceived negative on both scans and 2 had discordant evaluations on the standard end-diastolic image sets.The 3xMC+BC led to concordant test-retest findings in cases with originally discordant assessments on standard imaging.This finding shows that the complex motion patterns in combination with varying injection to scan delays not only affect the quantitative test-retest reproducibility but might also affect the clinical classification of lesions (Table 4, Figure 5).In addition, the TBR measure may vary due to differences in the reconstruction protocols.In a recent study, it was found that the TBR is also dependent on the postfiltering of the data and the number of iterations and subsets used [9].The use of timeof-flight and point-spread function corrections is expected to have a significant impact on the SUV assessment of the coronary lesions which often are of the same magnitude as the PET resolution [9].In the current study, we used an already established reconstruction protocol which was used in previous studies from our center [7,15].Further optimizations of the reconstruction protocols with a focus on improved reproducibility could be still possible.
Our findings show that a combination of comprehensive corrections for motion correction and injection to scan delays is critical for high reproducibility and consistent evaluation of lesions in coronary plaque imaging.To this end, the proposed correction techniques are applicable without the need for additional hardware or changes in the imaging protocols as motion and blood pool corrections, are performed using data obtained during standard acquisitions and can be performed retrospectively.The novel correction approaches presented in this study are directly applicable to ongoing clinical trials utilizing 18 F-NaF coronary plaque imaging: DIAMOND (NCT02110303) and PREFFIR (NCT02278211) [22,23].In principle, these motion correction methods may be also adapted to other coronary tracers and also to the 18 F-NaF PET imaging of patients with aortic stenosis [24].

Limitations
In this study, the number of cardiac and respiratory gates was limited to 4, due to reductions in count-statistics for each gated reconstruction.This approach, with a relatively limited number of gates, leads to increased intra-frame motion.Further improvement may be possible with larger number of gates or by correcting for motion either before or during image reconstruction [25], although such corrections are not implemented in current reconstruction toolboxes.In our study, the motion correction of the attenuation correction maps was not applied for each of the gated reconstructions.
This was not possible in the current setting as motion detection in projection space (sinogram space) is not easily transformed into motion in image-space.This limitation, however, is thought to have less impact on the reproducibility than performing no corrections for patient motion and respiratory motion, which is the current standard.
Future studies should address focus on optimization of motion correction of the attenuation correction for coronary PET imaging in addition to of the PET images.
Another limitation is the number of patients included in this protocol.However, this testretest study involves repeating of the complex CTA and PET protocol.Obtaining more extensive scan-rescan data with larger cohorts is currently not possible due to cost, ethics and radiation dose concerns.Partially mitigating this limitation, the total number of evaluated lesions was substantially larger than number of patients.Nevertheless, the results presented here were statistically significant, unequivocally demonstrating that the proposed corrections increase the coronary PET TBR reproducibility.

TABLES
Continuous variables reported as mean ± SD; categorical variables reported as n (%) Improvements of the TBR reproducibility were calculated against the standard.All incremental improvements (p-value) were tested using Pitman-Morgan tests, where p<0.05 were considered statistically significant).Continuous variables reported as mean ± SD.Following the co-registration, the MC images (MC, 2xMC and 3xMC) were obtained by averaging all the gated images.and gross patient motion corrected (3xMC).Significant reductions in the 95% confidence intervals (orange lines) were observed for all motion corrected datasets in comparison to the standard evaluation (Table 2).Standard = end-diastolic imaging.Significant reductions in the 95% confidence interval (orange lines) were observed for each incremental correction step (Table 5).BC = background blood pool correction.
Protocol PET/CT.Patients underwent 30-min list-mode PET-emission acquisitions approximately one hour after (66±9 min, range: 59-101 min) injection of 18 F-NaF (248±9 MBq).All patients were scanned in supine position with arms positioned above the head in a 128-slice Biograph mCT system (Siemens Healthineers, Knoxville, USA).A low-dose CT for attenuation correction was acquired immediately before the PET acquisition (120 kV, 50 mAs, 3-mm slice thickness).All patients were imaged with 3-lead electrocardiogram (cardiac gating), without the use of additional external markers for tracking of patient or respiratory motion.CT angiography.For anatomical localization of PET uptake, coronary CTA was performed immediately after the PET acquisition.The CTA imaging parameters including prospective gating, 330 milliseconds rotation time, body-mass index (BMI) dependent voltage (BMI <25, 100 kV; BMI ≥25, 120 kV), and tube-current time product of 160-245 mAs.Patients were administered beta-blockers (orally or intravenously) to achieve a target heart-rate of <60 beats/min.A BMI-dependent bolus-injection of contrast media

Figure 2 :
Figure 2: Reproducibility of coronary 18 F-NaF uptake measurement with motion correction.Bland-Altman plots of the target-to-background (TBR) evaluations for all the lesions with and without motion correction.(A) Standard, (B) cardiac motion corrected

Figure 3 :
Figure 3: Reproducibility of coronary 18 F-NaF uptake measurements and adjustments for blood pool clearance.Bland-Altman plots of the TBR assessment for all lesions observed in the study.Standard images without and with corrections for

Figure 4 :
Figure 4: Reproducibility of coronary 18 F-NaF uptake measurement in patients with substantial gross patient motion.Bland-Altman plots of the target-to-background (TBR) assessment for 18 F-NaF-avid lesions with gross patient motion >10 mm.Standard

Figure 5 :Patient 1
Figure 5: Test-rest coronary PET reproducibility before and after corrections.Patient 1 Patient with significant respiratory and gross patient motion during the first scan (10.3 mm) and a 20-minute difference in the injection-to-scan delay leading to discrepant