Anatomical validation of automatic respiratory motion correction for coronary 18F-Sodium Fluoride PET by expert measurements from 4D CT

Abstract

Background

Respiratory motion correction is of importance in studies of coronary plaques employing ¹⁸F-NaF; however, the validation of motion correction techniques mainly relies on indirect measures such as test-retest repeatability assessments. In this study, we aim to compare and, thus, validate the respiratory motion vector fields obtained by the positron emission tomography (PET) images obtained in FusionQuant application directly to the respiratory motion observed during 4-dimensional cine-computed tomography (CT) by an expert observer.

Purpose

To investigate the accuracy of the motion correction employed in a software (FusionQuant) used for evaluation of ¹⁸F-NaF PET studies by comparing the respiratory motion of the coronary plaques observed in PET to the respiratory motion observed in 4D cine CT images.

Methods

This study included twenty-three patients who undertook thoracic PET scans for the assessment of coronary plaques using ¹⁸F-Sodium Fluoride (¹⁸F-NaF). All patients underwent a 5-second cine-CT (4D-CT), a coronary CT angiography (CTA), and ¹⁸F-NaF PET. The 4D-CT and PET scan were reconstructed into 10 phases. Respiratory motion was estimated for the non-contrast visible coronary plaques using diffeomorphic registrations (PET) and compared to respiratory motion observed on 4D-CT. We report the PET motion vector fields obtained in the three principal axes in addition to the 3D motion. Statistical differences were examined using paired t-tests. Signal-to-Noise ratios (SNR) are reported for the single-phase images (end-expiratory phase) and for the motion-corrected image-series (employing the motion vector fields extracted during the diffeomorphic registrations).

Results

In total, 19 coronary plaques were considered in 16 patients. No statistical differences were observed for the maximum respiratory motion observed in x, y and the 3D motion fields (magnitude and direction) between the CT and PET (X direction: 4D CT = 2.5±1.5 mm, PET = 2.4±3.2mm; Y direction: 4D CT = 2.3± 1.9mm, PET = 0.7±2.9mm, 3D motion: 4D CT = 6.6±3.1mm, PET = 5.7±2.6mm, all p≥0.05). Significant differences in respiratory motion were observed in the systems’ Z direction: 4D CT = 4.9±3.4mm, PET = 2.3 ± 3.2mm, p=0.04. Significantly improved SNR is reported for the motion corrected images compared to the end-expiratory phase images (End-expiratory phase = 6.8±4.8, motion corrected = 12.2±4.5, p=0.001).

Conclusion

Similar respiratory motion was observed in two directions and 3D for coronary plaques on 4D CT as detected by automatic respiratory motion correction of coronary PET using FusionQuant. The respiratory motion correction technique significantly improved the SNR in the images.

1.1. Introduction

Positron Emission Tomography (PET) of the thoracic region is associated with quantificational inaccuracies due to blurring of the acquired PET data caused by cardiorespiratory motion. The detrimental impact of respiratory motion has been shown to translate the heart and non-small-cell lung cancers up to 30mm^1,2, while cardiac contraction might shift the right coronary artery up to 26mm¹. Correction for cardiorespiratory motions is of particular importance when imaging coronary artery uptake of a PET tracer. In this context, gating (cardiac, respiratory, and bulk patient motion and combinations thereof) of the acquired data has been suggested to reduce the detrimental impact of motion. However, the gated images only contain a fraction of the obtained counts obtained during the acquisition, leading to increased noise in the resulting PET images³. Motion correction of the gated images has been shown to improve both the signal-to-noise ratios (SNR), improved reproducibility, and the target-to-background ratios obtained in studies of coronary plaques^3–7.

The accuracy of the motion correction, however, relies on accurate tracking of organ movements during the scans. Several techniques have been proposed to detect the translations of the thoracic organs during both cardiac contraction and the respiratory cycle. For PET/Computed Tomography (CT) systems, cardiorespiratory motion tracing relies either on data-driven techniques employing only the acquired PET raw data (list data) or external markers such as 3-lead ECG-gating equipment^6,8–12. While these aspects are of importance to introduce motion correction into the clinical routine¹², the aforementioned measures do not directly validate the accuracy of motion compensation towards an anatomical standard (CT or Magnetic Resonance Imaging (MR)) in studies acquired in PET/CT systems. Moreover, it is common for respiratory tracing techniques to solely aim to detect the displacement of the heart and thoracic region.

Respiratory gating of the PET data has the disadvantage of only using a fraction of the acquired PET signal, thus, leading to higher noise levels in the resulting gated images¹³. Recently, several modern CT scanners support “cine” acquisition, which acquires multiple low-dose CT images during the duration of a complete respiratory cycle^14–16. In this study, we aimed to evaluate the accuracy of a motion correction protocol employed for ¹⁸F-NaF studies in FusionQuant by comparing the respiratory motion vector field (MVF) obtained during PET to PET image co-registration using a diffeomorphic registration algorithm against anatomical tracing of the same lesions derived during a single respiratory cycle using a 4D non-contrast cine-CT acquisition. This proof-of-concept study offers insight into the robustness of the motion detection and correction technique offered in studies of coronary plaques, which have low target-to-background ratios, and for the first time confirms that the motion obtained by PET respiratory motion correction algorithm estimates motion similar to what can be observed visually on anatomical CT images by expert observers.

1.2. Materials and Methods

1.2.1. Study population

This study comprised twenty-three patients who underwent ¹⁸F-NaF PET/CT examinations of the coronary arteries as part of a prospective longitudinal study evaluating the effect of PCSK9 inhibitors on coronary plaque (NCT03689946)¹¹. Among the inclusion criteria, the patients were required to have noncalcified coronary artery plaque (>440 mm3), as established by coronary CT angiography (CTA) obtained before the PET examination. The study was approved by the Cedars-Sinai Institutional Review Board.

1.2.2. Imaging protocol

In this study, three distinct image series were considered: a CTA series (1.2.2.1), a Cine-CT series (1.2.2.2), and an ¹⁸F-NaF PET series (1.2.2.3 and 1.2.2.4). The respiratory gated cine CT is not routinely obtained in the protocols and was obtained here only to validate respiratory motion obtained from gated PET data.

In clinical routine, the CT angiography and ¹⁸F-NaF PET images are acquired in a single-day, single imaging session. The routine assessment of the coronary plaques includes 3 steps; a co-registration of the PET and CTA images, extraction of the coronary centerlines (used for subsequent motion detection), and motion detection and correction step before the final assessments (1.2.2.5). For this purpose, the CTA serves both as an anatomical reference and to limit the motion correction strategy to only include the coronary arteries (1.2.2.5). Utilizing FusionQuant and the CTA-extracted coronary centerlines, a PET-to-PET motion correction is obtained using an optical flow strategy (diffeomorphic registration) (1.2.2.8). In this study, the cine-CT image series serve as validation to the MVF obtained in FusionQuant. The co-registration to the CTA images ensures a common baseline for the two compared image series for the anatomical guidance for the coronary plaque quantification (1.2.2.8).

1.2.2.1. CT angiography

Thirteen of the twenty-three patients had coronary CTA acquired on the same day as the ¹⁸F-NaF PET/CT examination, while the remaining ten patients had their CTA images acquired within 21 days (range=2–21 days). CTA images were acquired using prospective ECG-gating during an inspiratory breath-hold on a 192-slice Somatom Force Dual Source mCT system (Siemens Healthineers, Knoxville, TN, USA), with patients instructed to elevate their arms above the head. The CTA imaging parameters have been described previously¹¹. In brief end-systolic prospective gating was utilized. The dose administered to the patients from the CTA was 8mSv.

1.2.2.2. Cine-CT

On the same day as the ¹⁸F-NaF PET study, the patients underwent a non-contrast 4D-cine CT before the conventional imaging protocol, consisting of a standard low-dose CT attenuation correction map obtained immediately before the ¹⁸F-NaF emission acquisition. The cine-CT acquisition was a part of a study investigating the limitations of employing low-dose CT for attenuation correction purposes¹¹. Given the non-significant impact of using a respiratory averaged CT attenuation correction protocol¹¹, we have used the standard low-dose CT for attenuation correction purposes to stay in line with other studies evaluating coronary lesions. In general, we propose not to use the cine-CT acquisition for studies of coronary lesions as it adds no additional improvements in terms of attenuation correction and adds 1.4mSv (8.2% extra dose as compared to a ¹⁸F-NaF, low-dose CT for attenuation correction purposes and a CTA for a standard 70kg man) to the total dose given to the patient¹¹. Both the 4D-cine CT, low-dose CT attenuation correction map and ¹⁸F-NaF emission data were obtained in a hybrid PET/CT system (Discovery 710, GE Healthcare, Milwaukee, WI, USA). The 4D cine-CT acquisition protocol was acquired using the following settings: 120 kV, 10 mA, a field of view 50 cm, slice thickness 2.5 mm, and cine duration of 5 s as described previously¹¹. The respiratory tracking was obtained using a real-time position management system (RPM, Varian, Palo Alto). From the respiratory signal, ten respiratory-gated 4D-CT scans were reconstructed from the cine-CT scan using an amplitude-based reconstruction technique, with a voxel size of 0.98×0.98×2.5mm (Figure 1). The cine-CT accounted for a radiation dose of 1.4mSv, while the low-dose CT attenuation correction maps accounted for 1mSv.

Figure 1.

Coronal Respiratory motion was observed between the end-inspiratory and end-expiratory phases (blue and white lines, respectively), as obtained from a non-contrast respiratory-gated 4D-cine CT scan, shown as a single slice in a coronal orientation. The blue and white lines delineate the respiratory position of the diaphragm, liver, and heart obtained in the respective respiratory phases. Of note, the lesion shift (color-coded circles) was measured to be 5.5mm.

1.2.2.3. PET acquisition

All participants underwent ¹⁸F-NaF PET on a PET-CT scanner immediately after the 4D cine-CT acquisition. After the ¹⁸F-NaF target dose (250 MBq) administration, the participants rested in a quiet room for 180 min before the hybrid PET/CT acquisition (Figure 2). PET acquisitions were then performed and included respiratory tracing using a real-time position management system (RPM, Varian, Palo Alto) throughout a 30 min acquisition, used for the respiratory gating. The total dose administered to the patients from the PET accounted for 6.7mSv.

Figure 2.

Imaging protocol for the PET/CT acquisition. All patients rested for 180 minutes following injection of 18F-NaF before the PET/CT acquisition. Besides the PET emission acquisition, all patients had a cine-CT scan (4D-CT) and a CT scan for attenuation correction purposes.

CTAC = attenuation correction CT

1.2.2.4. PET reconstruction

10 respiratory gates (amplitude-based, truncated to contain data in the window of 10–90% of the range of respiratory signals) were reconstructed from the PET listmode dataset using a standard ordered expectation-maximization (OSEM) algorithm with time-of-flight and resolution recovery (point spread function correction)¹. Images were reconstructed using a standard low-dose CT attenuation correction map without any respiratory motion correction of the CT attenuation correction map. The PET reconstructions consisted of a 256 × 256 matrix (47 slices), using 4 iterations and 24 subsets (voxel size = 2.73×2.73×3.27mm)⁷.

1.2.2.5. PET to CT angiography and cine-CT image registration.

The PET images were co-registered to cine-CT and coronary CTA by rigid co-registration of PET images using five points of reference; sternum, vertebrae, blood pool in left and right ventricle, and the great vessels as described previously¹⁷. Of note, the respiratory gated ¹⁸F-NaF PET images were co-registered to the static CTA images using the end-expiratory PET phase.

1.2.2.6. Estimation of motion from 4D cine CT

Given the non-focal uptake patterns outside areas with active microcalcification, the respiratory motion was estimated only for lesions that could be identified on the gated non-contrast 4D-cine and PET reconstructions. The coronary lesions with identifiable calcification and visual uptake of ¹⁸F-NaF were segmented on both end-inspiratory and end-expiratory gated cine CT reconstructions using fixed-sized spheres (radius = 5mm). For each lesion and each respiratory phase on the CT, the sphere was inserted with the center covering the part of the lesion with the highest ¹⁸F-NaF uptake. We utilized CTA for anatomical validation of where the coronary plaques are located. Only lesions with concordant uptake of ¹⁸F-NaF and visible calcification on the cine-CT images were considered for this study. This was chosen to permit the comparison of the respiratory motion observed by an expert reader between the two modalities; this selection criterion is closely related to the usual selection criteria for single-lesion analyses where lesions with identifiable uptake on ¹⁸F-NaF PET and the CTA are segmented using spherical volumes of interest with 5mm radius^4,18,19. Co-registration of the gated CT images and PET images was necessary for all cases to ensure delineation of the lesions on the cine-CT and PET images and create a common baseline for the detection of the respiratory motion during the respective acquisitions. Utilizing the delineated calcified lesions, the respiratory motion was calculated from the offset in the center-of-mass of the coronary plaque segmentations obtained in the end-inspiratory and end-expiratory phases of the gated cine CT images.

1.2.2.7. FusionQuant

FusionQuant is a multimodality medical images viewer which permits the fusing of PET, CT, and MR images in any desired combination using multiplanar reconstruction. The program offers several analytical tools, hereunder the option of inserting different types of regions of interest (ROI) annotations to the images (lines, polygons, and spheres), and advanced processing tools such as image de-noising, registration, and motion correction are also integrated. FusionQuant is a stand-alone, research-only software tool that does not require any specific environment to work (Supplementary Figure 1).

1.2.2.8. Estimation of respiratory motion vector fields from PET

Detection and correction of respiratory motion obtained from the PET images were performed in a dedicated multimodality imaging viewer developed in our center (FusionQuant, Cedars-Sinai Medical Center)²⁰. To obtain local MVF for the calcified plaques only, from the automatically computed PET MVF (below), we placed 5mm radius spheres around the coronary plaques with visible uptake, with the center placed on the hottest voxel (highest standardized uptake value)⁴. Displacements of the calcified coronary plaques during the respiratory cycle were tracked using a PET-to-PET co-registration algorithm in FusionQuant described below. The respiratory motion was computed from the motion estimation of the co-registration algorithm. Of note, the gated PET images were supersampled to a spatial resolution of 0.1mm³ prior to the PET-to-PET image registration to ensure MVF with high spatial resolution.

FusionQuant co-registered the lesions from all of the gated PET images to the reference frame (end-expiratory phase) using a diffeomorphic registration algorithm^21,22. The deformable registration was used to register the coronary arteries, while only data within the segmented area were considered for further analyses. The motion correction was applied to the coronary artery tree, while only the inverse MVF within the segmented area were considered to evaluate the motion correction. In brief, the diffeomorphic registration is based on the optical flow (or image velocity) equation: $\overrightarrow{v.} \overrightarrow{\nabla \text{s}}=m-s$ with $\overrightarrow{v.}$ being the velocity, $\overrightarrow{\nabla \text{s}}$ the instantaneous velocity, m the intensity of an image M and S the intensity of an image S; the optical flow algorithm does not preserve mass. In brief, the velocity originates from optical flow terminology and refers to the displacement between two consecutive gates and does not preserve mass^23,24. This equation means that the intensity of a moving object is constant with time for small displacements. Optical flow behaves like deformable models based on attraction; the demons or attraction forces defined at every voxel, are iteratively computed, and the result of the application of a force during one iteration step is a displacement: $\overrightarrow{d}=-\overrightarrow{v}$ with $\overrightarrow{v}=\frac{\left(m-s\right)\overrightarrow{\nabla \text{s}}}{{\left(\overrightarrow{\nabla \text{s}}\right)}^2+{\left(m-s\right)}^2}$. The instantaneous optical flow or displacement for every point of the image are regularized at the end of each iteration by a gaussian function. This algorithm is computationally efficient compared to those based on elasticity, as it takes approximately 15 s to register 10 gates on a standard workstation.

The co-registration was performed within a region of interest of 10 mm around the coronary artery as derived from co-registered CTA images in Autoplaque software developed at Cedars-Sinai²⁵. The centerlines of the right, left circumflex, and left anterior descending coronary arteries were extracted for every patient in this manner. Using the region of interest around the coronary arteries, MVFs for the PET images were obtained during the diffeomorphic PET-to-PET respiratory gate co-registration process in FusionQuant (gated to reference frame registration)

From the co-registration process, the inverse MVFs were computed by subsampling the input MVF with a regular grid; using this grid (registering each respiratory frame to the reference respiratory frame) a spline was created using the landmarks. Using the grid and splines, the coordinates of the deformed point and the target landmarks are the negative of the displacement vectors. The spline is then used for regularly sampling the output space and recovering vector values for every single pixel. Of note, FusionQuant can be obtained through research agreements with Cedars-Sinai.

To compare the respiratory MVF derived by the above automatic algorithm and the respiratory motion obtained manually from cine-CT we focused on coronary lesions (landmark-based comparisons). In contrast, the respiratory motion observed in the gated CT images was defined as the changes in the center-of-mass of the segmented lesions obtained in end-inspiratory and end-expiratory phases. The expert-observed respiratory motion was calculated using a custom-made Matlab toolbox.

Common for both assessments, respiratory motion was reported as unidirectional measures (x, y, and z-direction) and in 3D ($\sqrt{x^2+y^2+z^2}$).

1.2.2.9. Signal-to-Noise ratio

Using the above-described motion vector field estimation, we introduced a motion correction of the gated PET images, thus, obtaining a motion-corrected PET image series. Signal-to-Noise ratios (SNR) are reported for single-phase and motion-corrected images as mean±standard deviation for grouped analyses of all lesions. The SNR was defined as (maximum lesion standard uptake value [SUV])/standard deviation(blood pool activity observed in the right atrium)²⁶.

1.2.3. Statistical Analysis

The vector displacements were tested for normality using the Shapiro–Wilk test. Continuous, normally distributed variables were presented as mean ±SD (standard deviation). Differences in the vectors were calculated using paired t-tests, using alpha levels of 0.05. Statistical analysis was performed using MedCalc Statistical Software version 19.1.7 (MedCalc Software bv, Ostend, Belgium). Boxplots were created using R 3.5.0.

1.3. Results

Twenty-three patients underwent ¹⁸F-NaF PET/CT examinations of the coronary arteries. A total of 19 calcified coronary plaques with focal ¹⁸F-NaF uptake were identified in 16 patients. In order to precisely measure the artery motion, only lesions with a calcified plaque were assessed. The remaining 7 patients with either no focal ¹⁸F-NaF uptake or no calcified lesions were not considered for the subsequent analyses. Baseline patient characteristics are shown in Table 1.

Table 1.

Baseline Patient Characteristics

Age±SD [years]	66±10y
Sex (Males)	21 (91%)
BMI (SD)	27±4
Hyperlipidemia	17 (74%)
Hypertension	14 (61%)
Diabetes	5 (22%)
Smoker/ex-smoker	7 (30%)
Angina (%)	8 (35%)
ACS (%)	7 (30%)
Previous PCI /CABG (%)	5 (20%)
Previous TIA/Stroke (%)	1 (4%)
Diabetes (%)	10 (31%)
Medication
Aspirin	19 (82%)
Statin	23 (100%)
ACE/ARB	9 (39%)
Quantitative plaque burden median [iqr]
Total plaque volume [mm3]	837 [620–1055]
Total non-calcified plaque volume [mm3]	711 [550–859]
Total calcified plaque volume [mm3]	102 [43–205]
Agatston score [AU]	350 [0–1200]

Continuous variables reported as mean ± SD or as median and [iqr]; categorical variables reported as n (%).

1.3.1. Respiratory motion

Unidirectional and 3D respiratory MFV were calculated for the PET images, while the corresponding respiratory motion were calculated for the gated cine-CT images. Regarding respiratory shifts, most significant motion was observed for the z-direction with average shifts of (4D CT = 4.9±3.4mm, PET = 2.3 ± 3.2mm, p=0.04), compared to (x: 4D CT = 2.5±1.5mm, PET = 2.4 ± 3.2mm, p=0.82, y: 4D CT = 2.3±1.9mm, PET = 0.7 ± 2.9mm, p=0.07) (Figure 3). The overall 3D motion was similar when comparing 4D-CT and PET images (4D-CT = 6.6±3.1mm, PET = 5.7±2.6mm, p=0.31) (Figure 3).

Figure 3.

Respiratory motion observed from end-inspiratory and end-expiratory phases from respiratory-gated CT and PET images. The respiratory motion shifts from the individual directions are shown in A-C, while the 3D motion is shown in D.

Figure 4 shows the differences in motion obtained from 4D-CT and PET, respectively. On average, the differences were 0.1 mm in the x-direction, 1.6 mm in the y-direction, 2.6 mm in the z-direction, and a difference of 0.9 mm for the 3D vectors. Corresponding % wise differences (mean ± standard deviation) were observed for the motion patterns: 8.6±90.2 %(x-direction), −7.6±84.5% (y-direction), −30.5±93.0% (z-direction) and −0.5:50.1% (3D motion), respectively.

Figure 4.

Bias in the measured motion between PET and 4D-CT. Biases of less than 5mm were observed for most of the patients.

A case example of respiratory motion observed in both 4D-CT images and the PET/CTA images are shown in Figure 5. Figure 6 shows a case example of the absolute motion vector map obtained from respiratory-gated PET in FusionQuant.

Figure 5.

Case example of respiratory motion observed for a lesion in the proximal left anterior descending artery, as observed on both CTA, 4D-CT, and 18F-NaF PET. A) shows the respiratory motion of the myocardium during the cine-CT, shown as a shift in the anterior wall (blue outline = end-inspiratory position of the myocardium, white outline = end-expiratory position of the myocardium). Furthermore, the respiratory motion of the coronary lesion is shown by the blue and white circles connected with the arrow. B) shows the PET and contrast CTA images fused. In brief, the big light grey areas show the left ventricle and atrium/aortic arch; the darker grey areas show the myocardial tissue, and the smaller bright areas on the CT image show the LAD territory, with the overlayed PET (yellow activity) indicating the location of the coronary lesion with active microcalcification. Overlaid on the PET/CTA fusion is shown the motion vector fields obtained during the PET-to-PET image registration; the shift for both scans was measured to be 5.5mm.

Figure 6.

Case example of respiratory motion observed for a lesion in the proximal left artery descending, as overlayed on the CTA image (in the background). Respiratory MVF were obtained from PET within a 10mm radius of the coronary arteries (green volume of interest). PET MVF are shown using the color map. The respiratory motion estimated from 4D-CT is shown by the arrow. The average 3D motion observed in the lesion was 5.3mm (PET motion fields), while the corresponding 3D motion observed on the 4D-CT was 7.1mm.

MVF: motion vector field, CTA: CT angiography

Employing the motion detection and correction technique led to significant improvements in the SNR (End-expiratory phase = 6.8±4.8, motion-corrected = 12.2±4.5, p=0.001).

1.4. Discussion

Due to the relatively small size of the coronary arteries and their motion during respiratory and cardiac cycles, correction for motion is necessary for cardiac PET imaging of the coronary arteries. In this study, we evaluated the accuracy of PET MVFs obtained using diffeomorphic registrations with respiratory shifts obtained from 4D respiratory-gated CT images derived from a cine-CT acquisition. Our main finding was that respiratory motion obtained from PET and the respiratory shift obtained from the respiratory-gated CT images were comparable. Motion correction of the PET images using the MVF’s improved the SNR, thus suggesting that motion correction obtained for PET images diffeomorphic registrations provide accurate motion correction when correcting for respiratory motion. To our knowledge, this is the first study evaluating the direct accuracy of displacements observed in PET to anatomically identifiable displacements obtained from 4D cine-CT scans. Although the impact of the motion detection was focused on coronary lesions with ¹⁸F-NaF uptake, the motion detection and correction using diffeomorphic co-registration algorithms are also applicable for scans where anatomical uptake patterns can be observed for the PET images, such as in myocardial perfusion and metabolism imaging protocols as well as in oncological studies of the diaphragm and lungs. The motion correction is currently implemented in FusionQuant, a research software that can be obtained through a research agreement with Cedars-Sinai and is currently used at several research sites. Therefore, many centers focusing on the assessment of ¹⁸F-NaF cardiovascular PET or other plaque tracers (⁶⁸Ga Dotate, FDG, etc.) (in coronary plaques, aortic valves, aortic arch, and arteries in the head/neck and extremities) can use the registration algorithm described in this study.

¹⁸F-NaF has been proven to provide assessments of active calcification (microcalcification) across a range of cardiovascular conditions^{18,19,27–30}. Most studies currently employ either end-diastolic or cardiac motion-corrected images, which have been shown to have reduced signal-to-noise ratios³¹, reduced reproducibility in test-retest assessments⁹, and potentially wrong lesion classifications when respiratory and patient motion is not included in the motion correction protocol⁵. The inherent noise in the gated images increases with the number of gates employed in the reconstruction protocol. While the noise may affect the accuracy of the motion detection in the images, the uncertainty of the accurate co-registration is expected to be minimal as only local motion patterns are taken into consideration. Supporting this hypothesis are the findings in the current study, where similar respiratory motion was observed for the 4D cine CT and PET images in the x and y direction, while a slight reduction was reported for the PET MVF in the z-direction (Figure 3). The reduction in the MVF, however, did not exceed the voxel sizes of the CT images (0.98×0.98×2.5mm) or the PET images (2.73×2.73×3.27mm). The discrepancy in the voxel resolution of the images might add to the bias in the respiratory motion obtained from the two scans as the MVF observed for the PET system is of magnitude with the intrinsic resolution of the PET system (Figure 4). In this context, this study demonstrates that the optical flow strategy accurately depicts motion in radionuclide imaging applications, even when applied to noisy image series. Further, this suggests that the accuracy in the motion correction is driven by the spatial resolution of the PET system, in agreement with the Nyquist sampling theorem. The differences in spatial resolution of the two systems are further complicated by differences in the partial volume effects obtained for the PET and CT scans, which will affect the accuracy of the MVF estimates. Further, significantly improved SNR was reported following respiratory motion correction, thus, suggesting that the accuracy of the MVF was not affected by the discordance observed for the respiratory motion in the z-direction. Therefore, the findings in this study indicate that the algorithm successfully corrects for the respiratory motion of the coronary arteries despite minor differences in the motion observed for the 4D-CT and PET-PET approaches. Related to the comparison of the MVF and the respiratory motion observed on the cine-CT images; the ideal comparison of the motion observed for the two techniques would have included a direct comparison of the diffeomorphic registrations of the gated CT images. However, the optical flow algorithm has not yet been validated for the gated CT images, and would, thus, not have served as an objective metric. Therefore, we opted to utilize as a “gold standard” the current clinical judgment of the motion, which is not dependent on any possible algorithm issues, and it definitively and independently established the amount of motion on CT - as currently considered during the clinical scans. Of note, the accurate respiratory motion correction is of great importance in both research and clinical practice. Test-retest repeatability often can vary when motion correction is not applied⁹; however, by limiting the test-retest variation, it is possible to set up diagnostic criteria, such as threshold values for the detection of vulnerable plaques.

Most studies on respiratory motion detection and correction have focused on targeting the detection of the respiratory signal instead of tracking the actual motion observed in the images. One limitation of this technique, using external markers, is that the respiratory trace and underlying motion do not necessarily match, given the non-linear changes observed on the chest. In this study, we evaluated the underlying motion using the same infrared tracking device to track respiratory motion for both PET and CT, which was used to generate 10 phases for both systems, allowing for a direct comparison of the motion. Although the motion correction relied on MVFs obtained using external markers for the detection of the respiratory trace, recent studies from these systems have shown a great benefit of using data-driven (image-driven) motion detection and correction techniques^32–34. Regarding the impact of the motion correction, in this study, we focused on the SNR metric to report the improvements observed in the motion-corrected images. While the SNR may not be the best metric to assess improvement, as even minor corrections for motion will improve this metric, it provides an insight into the impact of the motion correction. With the improvements in the SNR in combination with the harmonic findings for respiratory motion, we believe that the motion correction provides significantly improved images for clinical reading.

In the context of attenuation correction, for this study, we relied on the standard low-dose CT images for attenuation correction purposes. From the cine-CT acquisitions, it is possible to create respiratory averaged CT attenuation correction maps. In a previous study from our center, we have demonstrated that the use of respiratory averaged CT attenuation maps does not affect the quantitative accuracy of coronary lesions¹¹. Therefore, we do not see the use of a single-phase CT attenuation correction as a limitation of this study.

Our study has some other limitations. The analysis is limited by the discrepancy in the duration of the PET and the cine-CT system, which can cause changes in the respiratory depth¹⁰. The 4D cine-CT scans were acquired over 5 seconds, compared to the 30 min long PET scans. The cine-CTs were only acquired for 5 seconds which may not cover a full respiratory cycle. Precautions were taken to limit the risks of not covering the full respiratory cycle. The patients were placed on the bed and asked to breathe normally for approximately 30s before the cine-CT acquisition to minimize the risks of lengthy breathing cycles (>5 seconds). Despite these precautions, we had 2 cases (excluded from the study) who had a respiratory cycle exceeding 5 seconds. However, the disease burden and advanced age of study participants led to relatively short respiratory cycles in our patients (within the 5 seconds acquisition window).

This study has additional limitations. The study comprised 16 patients. Despite the small number of participants, we were able to evaluate the respiratory motion obtained from both PET and CT in 19 lesions, with comparable motion observed for both approaches. Also, relating to the modest number of study participants and the limited number of lesions, we employed an alpha-level of 0.05 for the test-statistics. This may mask some differences between the MVF obtained from PET and the respiratory motion observed on CT. Another limitation in terms of the PET-derived MVF is that they only are estimated locally in the coronary arteries and, thus, do not provide insight into the full extent of motion observed in the entire PET image. Given that only local motion correction was applied, no problems were reported for the MVFs. A general co-registration of the PET images may improve the image quality outside the coronary lesions, for instance, in studies of aortic valves where motion also has a detrimental impact³⁵. However, introducing a global motion correction technique may add both computing time and the risks of problems in image co-registrations, given the noisy nature of ¹⁸F-NaF PET imaging. The noise is especially of concern in the liver/lung area, which also is affected by respiratory motion. Lastly, this study was performed in a single center using a PET/CT system from one vendor only. A larger multicenter study incorporating systems from multiple vendors would be required to confirm our findings.

1.5. Conclusion

In a cohort of thirteen patients, we found that respiratory motion estimated from respiratory motion compensation from PET alone by an automated algorithm employed in FusionQuant is comparable to the motion obtained from respiratory-gated CT images in 2 principal directions (x and y) and in 3D. Motion correction of the images significantly improved the SNR compared to the end-expiratory phase images.

Abbreviations

CTAC

CT attenuation map

CT

Computed Tomography

MR

Magnetic Resonance (Imaging)

¹⁸F-NaF

¹⁸F-sodium fluoride

PET

positron emission tomography

MVF

Motion Vector Fields

CTA

computed tomography angiography

VOI

volume of interest

SNR

Signal to noise ratio