Comparing approaches to correct for respiratory motion in NH3 PET-CT cardiac perfusion imaging.

AIM: Respiratory motion affects cardiac PET-computed tomography (CT) imaging by reducing attenuation correction (AC) accuracy and by introducing blur. The aim of this study was to compare three approaches for reducing motion-induced AC errors and evaluate the inclusion of respiratory motion correction.
MATERIALS AND METHODS: AC with a helical CT was compared with averaged cine and gated cine CT, as well as with a pseudo-gated CT, which was produced by applying PET-derived motion fields to the helical CT. Data-driven gating was used to produce respiratory-gated PET and CT images, and 60 NH3 PET scans were attenuation corrected with each of the CTs. Respiratory motion correction was applied to the gated and pseudo-gated attenuation-corrected PET images.
RESULTS: Anterior and lateral wall intensity measured in attenuation-corrected PET images generally increased when PET-CT alignment improved and decreased when alignment degraded. On average, all methods improved PET-CT liver and cardiac alignment, and increased anterior wall intensity by more than 10% in 36, 33 and 25 cases for the averaged, gated and pseudo-gated CTAC PET images, respectively. However, cases were found where alignment worsened and severe artefacts resulted. This occurred in more cases and to a greater extent for the averaged and gated CT, where the anterior wall intensity reduced by more than 10% in 21 and 24 cases, respectively, compared with six cases for the pseudo-gated CT. Application of respiratory motion correction increased the average anterior and inferior wall intensity, but only 13% of cases increased by more than 10%.
CONCLUSION: All methods improved average respiratory-induced AC errors; however, some severe artefacts were produced. The pseudo-gated CT was found to be the most robust method.

Introduction

In dual-modality PET-computed tomography (CT) imaging, respiratory motion can produce two image-degrading effects. PET images are usually acquired over several respiratory cycles, resulting in motion-averaged images that contain blurring. The second effect is a result of the different temporal resolutions of PET and CT. In contrast to the long PET acquisition time, modern multislice CT scanners acquire helical CT (HCT) scans quickly (the lungs can be acquired in under 2 s), and each slice represents a snapshot of the respiratory cycle. Resulting spatial inconsistencies between PET and CT can lead to attenuation correction (AC) errors, particularly when regions of soft tissue are corrected with lower attenuation coefficients of lung 1.

In cardiac PET imaging, respiratory motion blurring in PET has been shown to create regions of significant heterogeneity in the myocardium 2. In cardiac PET-CT imaging, AC errors resulting from respiratory motion have been identified as a source of artefacts that can resemble false defects 3, which can be severe when the PET-CT mismatch of the myocardium exceeds 8 mm 4,5. Artefacts typically appear as reduced uptake in the anterior and lateral walls where a PET-CT misalignment of the myocardium exists; however, bias in septal 6 and inferior 5,7 walls can also occur from PET-CT misalignment of the liver.

Several approaches for reducing the effect of respiratory motion in cardiac PET-CT imaging exist. These are of varying degrees of complexity, and most can be generalized into three main categories. The first produces an AC map, which more closely matches the average respiratory position represented in the PET data. This can be achieved with breath-hold techniques 8–10, manually or automatically translating the CT to match the PET 4,11, acquiring a slow-rotation HCT 5,12, or acquiring a cine CT and averaging over the temporal dimension 13–15. These approaches reduce respiratory-induced AC errors; however, motion blurring is not addressed.

In the second category, PET and cine CT images are respiratory gated, enabling matched AC in which each PET gate is attenuation-corrected with CT data from the matching point in the respiratory cycle 16–18. This reduces AC errors that result from PET-CT misalignment. In addition, each PET gate produces a near motion-free image with reduced respiration motion blur. However, gating PET divides the acquired counts into several separate images, and this results in increased noise, which compromises the benefits of gating 19. Summing the attenuation-corrected images produces a single image with reduced AC errors and with noise similar to the original, noncorrected PET image; however, motion blurring is reintroduced.

The third category reduces both motion blur and AC errors while maintaining image noise by motion correcting and recombining the matched attenuation-corrected PET data. This can be achieved either by reconstructing and then adding the gated images 20,21 or by including respiratory motion correction into iterative reconstruction methods 22–24.

As an alternative to using a gated cine CT for matched AC, a pseudo-gated CT can be generated by applying motion fields derived from gated PET images to a three-dimensional (3D) HCT volume. This approach has been demonstrated with thoracic images using nonlinear optical flow 25 and B-spline 26 registration. In cardiac PET-CT imaging, psuedo-gated CTs have been generated using rigid registration 27 and with respiratory models 28. Pseudo-gated CTs offer an advantage over gated cine by reducing the dose; a 16 cm cine CT over the chest acquired using 140 kV, 10 mA and 5.5 s cine duration imposes a dose of ∼3 mSv, in contrast to ∼0.5 mSv resulting from an HCT of 8 mAs and 140 kV over the same region. In addition, the amplitude and period of respiration can vary from cycle to cycle 29–32, which can cause artefacts in gated cine CT 33. Pseudo-gated CTs avoid this; however, they do rely on an artefact-free HCT that matches the respiratory position of one PET gate.

The aim of this study was to compare approaches that reduce cardiac NH3 PET-CT misalignment-induced AC errors and motion-induced PET blurring. In particular, we determined which approaches are more likely to create, rather than reduce, respiratory motion artefacts in NH3 cardiac PET-CT images. First, we evaluated approaches that reduce only AC artefacts: averaged cine CT (ACT), matched AC using gated cine CT and matched AC using pseudo-gated CT. These were compared with the standard approach of HCT-based AC. Following this, we evaluated the inclusion of respiratory motion correction with the matched AC images.

Materials and methods

Rest and stress NH3 myocardial perfusion scans from 37 patients with known or suspected coronary artery disease were included in this study, and images were acquired on a GE Discovery VCT PET/CT scanner (General Electric Medical Systems, Waukesha, Wisconsin, USA). First, an HCT scan was acquired over 150 mm using 140 kVp, 20 mA, 0.4 s/rotation and a pitch of 1.375 : 1, and was reconstructed to 2.5-mm-thick slices. Following the HCT, patients were injected with ∼550 MBq of 13NH3, and a 2D PET image was acquired for 26 min in listmode. The initial rapidly changing portion of NH3 scans is not included in the relative assessment of myocardial perfusion 34, and for this work the 20 min of PET data obtained 6 min after injection were used. A cine CT image was then acquired at 140 kVp, 10 mA, 0.4 s/rotation, 5.5 s cine duration and 40 mm collimation, and was reconstructed to 2.5 mm contiguous slices. ECG gating was not used for PET or CT imaging.

Voluntary patient movement was determined by measuring the displacement of the spine between the helical scan and the cine CT scan, which were on average 30(±4) min apart. In 14 cases, movement greater than 5 mm (approximately the resolution of PET) was measured, and these data sets were excluded from this study. The remaining 60 individual image data sets were analysed further. All patients consented to take part in the study, which was approved by a national ethics committee.

Respiratory gating

PET and cine CT acquisitions were respiratory gated using data-driven gating. The gating method used, which has been validated against a hardware-based gating method, derives a respiratory signal by observing the intensity change in the pixels, which are subject to respiratory motion throughout the acquisition 35,36. PET and CT data were gated into eight gates. Variably sized amplitude gating was used to gate the PET data, in which the size (amplitude range) of each gate was adjusted such that an equal amount of acquisition time was assigned to each gate 37. This ensured similar noise statistics in each sinogram gate.

Image registration

All nonrigid registrations used to create the pseudo-gated CT and for respiratory motion correction of PET were performed with NiftyReg version 1.3 (University College London, London, UK) 38. This implements the B-spline-based free-form deformation method developed by Rueckert et al. 39, which uses normalized mutual information to determine image similarity at the voxel level. A 10 mm3 final grid, 10 mm source and target filter, and a bending energy weight of 0.05 were used.

Attenuation-corrected PET data sets

Six attenuation-corrected PET image volumes were created for each scan, and these are summarized in Table 1. In method 1, the PET image attenuation-corrected with the HCT, ACHCT, was used as the baseline for comparison. Methods 2–4 reduce AC artefacts only, and methods 5 and 6 reduce both AC artefacts and motion blur in PET by including registration. All PET reconstructions were performed with filtered back-projection using a 5.7 mm Hamming filter, 70 cm diameter and 293×293 slices.

Table 1

Summary of the six AC PET images created for each study

ACHCT: The HCT was interpolated in the axial direction to match the positions of the PET slices before converting to CTAC. Gated PET sinograms were summed, attenuation-corrected, and reconstructed to create the ACHCT PET volume.

ACACT: All CT slices at each slice location of the 4D cine CT were averaged to create a single CT volume, which was interpolated in the axial dimension to match the PET position. Gated PET sinogram gates were summed, attenuation-corrected with the ACT and reconstructed to create the ACACT PET volume.

ACGCT: Each PET sinogram gate was attenuation-corrected with the CT gate with matching gate number. Gated attenuation-corrected PET data were then reconstructed and summed to create the single ACGCT PET volume.

ACpseudo: A pseudo-cine CTAC was created by applying the motion fields derived from the non-attenuation-corrected (NAC) PET images to the helical CTAC image. First, the NAC PET gate that spatially matched the helical CTAC was identified by locating the superior edge of the liver in both data sets (described below), and the PET gate with the closest liver position to the helical CTAC was defined as the matching gate. This matching NAC PET gate was registered to each remaining NAC PET gate, creating seven transformations that were then applied to the helical CTAC. This produced a CTAC for each respiratory position represented by the gated PET series, effecting the pseudo-gated CT sequence. PET sinograms were attenuation-corrected with the matching pseudo-gated CT gate, reconstructed, and summed to create the single ACpseudo PET volume.

ACRegGCT: Gated PET sinograms were attenuation-corrected with matching CT gates as for method 3, and individually reconstructed. All PET image gates were registered to the expiration gate, which was assumed to be the gate with the most superior centre of mass. To avoid compounding the effects of any remaining spatial misalignment between PET and AC maps, motion correction transformations were derived from gated, NAC PET images, and the transformations were applied to the attenuation-corrected images. Following registration, all attenuation-corrected PET gates were summed.

ACRegpseudo: The same procedure as applied for method 5 was carried out, except for the use of pseudo-gated CT rather than gated cine CT.

Evaluation

The six respiratory motion correction approaches (methods 1–6) were evaluated in two separate stages. First, the different AC methods (methods 1–4) were compared using two metrics: PET-CT alignment and relative difference in myocardial wall intensities in attenuation-corrected short-axis (SA) PET images. In the second stage, the effects of implementing registration (methods 5 and 6) were evaluated.

Attenuation correction

First, alignment between the NAC PET images and each CTAC was assessed to compare the four approaches of AC (methods 1–4). Liver and myocardium positions in each NAC PET and CTAC gate were determined from profiles over each organ (described below). Alignment was defined as the absolute difference between NAC PET and CT positions averaged over all gates. For the gated and pseudo-gated CTs, each NAC PET gate was compared with the matching CTAC gate, and for the averaged cine and HCT each NAC PET gate was compared with the single static CT.

Second, the four AC methods were compared by measuring the myocardium intensity in the attenuation-corrected images ACHCT, ACACT, ACGCT and ACpseudo. The AC PET image volumes were rotated to create SA slices, which were smoothed with a 3D Gaussian 4.8 mm FWHM filter to reduce image noise and allow for direct comparison of volumes. Two profiles, one towards the base and one towards the apex, were manually defined on each of the anterior, inferior, lateral and septal walls of the SA slices. For each acquisition, one set of profiles was defined and copied onto all AC image data sets. The peak of each profile was used to indicate cardiac wall intensity.

As the true myocardial activity is unknown, relative change in myocardial intensity was used to evaluate AC methods 2–4, using the ACHCT images as the baseline for comparison. When poor PET-CT alignment exists, the anterior wall of the myocardium in PET can correspond to the lung in the CT, and this can result in undercorrection for attenuation and reduced activity in the attenuation-corrected PET image. Replacing this CT with a CT that improves alignment is therefore expected to increase anterior wall intensity in the AC PET images. Conversely, a decrease in anterior wall intensity in AC PET would result if the replacement CT worsens PET-CT alignment. A positive correlation between the change in PET-CT alignment and change in anterior wall activity is therefore expected.

Respiratory motion correction

The range of liver and cardiac motion was determined by measuring the liver and myocardial edge (described below) on the NAC-gated PET image series on both the motion corrected and non-motion-corrected images. Accuracy of respiratory motion correction was indicated by the reduction in motion range.

The effect of applying respiratory motion correction was also determined by evaluating the change in myocardial wall intensity in attenuation-corrected PET images. Profiles were defined over the myocardium on SA images ACRegGCT and ACRegpseudo, and on the nonregistered SA images (rigid body registration was first used to align the ACReg images with the AC images such that the profile locations copied onto all data sets represented corresponding anatomical positions). ACGCT and ACpseudo, and the change in cardiac wall intensity, described the effect of applying respiratory motion correction. As respiratory motion can blur the myocardial wall and effectively reduce the wall intensity, an increase in wall intensity is expected when motion blurring is reduced.

Liver and myocardial edge

The location of the liver and myocardium along the z-axis in PET and CT images was necessary to match the HCT with the gated PET (to produce the pseudo-gated CT) and to evaluate PET-CT alignment. To determine the position of these organs, profiles of the superior edges of the liver and myocardium were obtained by projecting a volume of interest defined over the organ edge onto the z-axis. To reduce profile noise in the PET images, a further 4 mm Gaussian filter was applied before defining the volumes of interest.

A straight line was fitted to the central segment of the profile, which spanned 20–80% of the profile range. The point where the fitted line intercepted 50% of the profile range defined the approximate location of the organ edge along the z-axis. Figure 1 illustrates the liver edge definition for PET and CTAC slices from the same patient. The displacement between the liver positions in PET and CTAC is highlighted.

Fig. 1

PET-CT alignment of the liver. A profile was defined along the z-axis over the liver edge in both data sets, and a straight line was fitted to the section ranging from 0.8 to 0.2 of the intensity normalized profile. Alignment was defined as the z-axis separation between the two straight lines at y=0.5. CT, computed tomography; NAC, non-attenuation-corrected; VOI, volume of interest.

Results

Averaged, gated and pseudo-gated CTs were generated for all acquisitions. In 45 of the 60 scans, the pseudo-gated CT produced a larger range of myocardium respiratory motion compared with the gated cine. On average, the range of liver and myocardium respiratory motion measured in the pseudo-gated CT was 13.1 and 6.9 mm, respectively, significantly greater than the average motion measured on the gated cine, which was 7.6 and 4.9 mm, respectively.

Attenuation correction

PET-CT alignment

On observing the overall PET-CT alignment for all scans, we found that the pseudo-gated CT produced the best liver and cardiac alignment, with a mean alignment of 1.9 and 5.4 mm, respectively. The HCT produced the worst liver and cardiac alignment, with 5.2 and 6.5 mm, respectively.

Compared with the HCT, the ACT, GCT and pseudo-gated CT all improved the mean PET-CT alignment of the liver and myocardium, as shown in Table 2. Only the pseudo-gated CT demonstrated a significant improvement of both the liver and the myocardium. All three cine CTs also produced cases in which cardiac wall alignment was worse when compared with the HCT. The pseudo-gated CT produced the smallest degrading effect, up to 3.2 mm, whereas the alignment worsened by up to 9.6 and 11.2 mm for the average and gated cine CTs, respectively.

Table 2

Mean±SD improvement (mm) in liver and myocardial wall PET-CT alignment (no motion correction) using the HCT as a baseline for comparison

Cardiac wall intensity

Using the HCT as a baseline for comparison, the SA PET images ACACT, ACGCT and ACpseudo were compared with the ACHCT images to observe the relative change in myocardial wall intensity. The PET images of two example patients are shown in Fig. 2. In Fig. 2a, the ACT, GCT and pseudo-gated CT are all seen to improve the alignment when compared with the HCT. An anterior apical wall defect apparent in the ACHCT PET image is not present in the ACACT, ACGCT and ACpseudo PET images. For this case, the mean liver and cardiac wall PET-CT misalignment was reduced from 10.6 and 7.9 mm, respectively, when the HCT was used to 2.7 and 3.0 mm for the GCT and 4.7 and 0.6 mm for the pseudo-gated CT, respectively.

Fig. 2

(a) Example 1. (b) Example 2. Bullseye and horizontal long-axis images of two example cases. (a) An artefact that resembles a defect on the anterior wall is seen on the ACHCT PET images, which results from a PET-CT spatial mismatch. This artefact is corrected in ACACT, ACGCT and ACpseudo images from improved PET-CT alignment. (b) Good PET-CT alignment exists for the ACHCT and ACpseudo data sets, and no defect is seen. For the ACACT and ACGCT images, worse PET-CT alignment produces an artefact that appears as an anterior wall defect. AC, attenuation correction; ACT, averaged cine CT; CT, computed tomography; HCT, helical CT; NAC, non-attenuation-corrected.

In the second example, Fig. 2b, the ACT and GCT both produced worse cardiac PET-CT alignment (9.9 and 11.5 mm, respectively) compared with the HCT (0.3 mm) and pseudo-gated CT (1.6 mm). A spatial PET-CT misalignment for ACT and GCT results in an artefact that appears as an anterior wall defect and is seen in the ACGCT PET image but not in the ACHCT PET image. A 37% decrease in anterior wall intensity was found with the ACGCT PET data. The ACpseudo PET images were not visually different from the ACHCT images.

The mean change in intensity of the SA walls for all patients is shown in Table 3. In Fig. 3, the change in anterior and lateral cardiac wall intensities is shown as histograms, in which more of a positive bias can be seen for the ACpseudo PET images. The ACACT and ACGCT PET images showed more cases of a large decrease in intensity. A decrease in anterior wall intensity of more than 10% was seen in 21 and 24 cases in the ACACT and ACGCT PET images, respectively, compared with only six cases in the ACpseudo images. However, slightly fewer cases were found when the anterior wall increased by more than 10% in the ACpseudo images: 25 cases compared with 36 and 33 cases in the ACACT and ACGCT images, respectively.

Table 3

Mean±SD change (%) in SA wall intensities using ACHCT as a baseline for comparison

Fig. 3

The percentage change in cardiac (a) anterior and (b) lateral wall intensities in PET images when the helical CTAC was replaced with the average, gated and pseudo-gated CTACs. More of a positive bias is seen with the ACpseudo images. AC, attenuation correction; ACT, averaged cine CT.

The septal and inferior walls were less affected; however, an increase of up to 22% and a decrease of up to 39% were found. No correlation was found between the change in inferior wall intensity and the change in cardiac wall alignment; however, a correlation was found with liver alignment (P<0.001). Similarly for the septal wall, a higher correlation with wall intensity change was found with liver alignment than with cardiac alignment. When the pseudo-gated CT improved alignment by reducing a superior displacement of the liver, the inferior and septal wall intensity decreased by more than 10% in 13 cases, suggesting overcorrection of attenuation in the septal and inferior walls by superior displacement of the liver with the HCTs.

The change in anterior wall intensity was well correlated with the change in PET-CT cardiac alignment (ρ=0.72, P<0.001), and, in general, the anterior wall intensity increased when alignment improved and decreased when the alignment was poor. Figure 4 shows the percentage change in anterior wall intensity and the change in alignment for each anterior wall SA measurement.

Fig. 4

The change in both PET-CT alignment and anterior wall intensity that results from replacing the HCT with the ACT, GCT or pseudo-gated CT. Generally, an improvement in alignment (a positive change) creates an increase in anterior wall intensity, and a worse alignment (a negative change) produces a decrease in anterior wall intensity. The solid diagonal line indicates linear fit of all data (y=0.14x+0.21). ACT, averaged cine CT; CT, computed tomography; HCT, helical CT.

Exceptional cases were noted on the gated and averaged cine data sets, in which an increase in anterior wall intensity of more than 10% resulted from a worse PET-CT alignment. In these cases, motion artefacts in the gated and average cine CTs produced a spatial discontinuity along the anterior wall, and the alignment measurement did not accurately represent the PET-CT displacement at the apical section of the anterior wall. In one case, an improvement in PET-CT alignment resulted in a decrease in anterior wall intensity of ∼20%. In this case, the myocardium position in the HCT was superior to the position in PET. Although the absolute PET-CT alignment improved when the ACT and GCT were used, the myocardium position in these CTs was up to 10 mm inferior to the PET position in some gates, resulting in an undercorrection of the anterior wall.

Respiratory motion correction

The range of respiratory motion was measured on the gated NAC PET images before and after respiratory motion correction. On average, the range of liver motion reduced from 12.2 to 1.0 mm, and the cardiac wall respiratory motion range reduced from 8.3 to 0.8 mm.

Application of respiratory motion correction to the ACGCT and ACpseudo PET images significantly increased the intensity of all SA walls (P<0.05), with the largest change seen in the anterior and inferior walls, as shown in Table 4. The change in wall intensity was significantly correlated with total respiratory motion range for the anterior and inferior walls (P<0.001); however, no correlation was found for the septal and lateral walls. This is demonstrated in Fig. 5, which shows the change in cardiac wall intensities measured on ACRegGCT and ACRegpseudo images, with respect to the total respiratory-induced displacement range measured on NAC PET images.

Table 4

Average change (%) in SA wall intensities resulting from respiratory motion correction

Fig. 5

For the anterior (a) and inferior (b) walls, a significant (P<0.05) correlation was found between the effect of respiratory motion correction (the change in profile maximum measured on ACRegGCT and ACRegpseudo) and the range of respiratory-induced motion measured on NAC-gated PET images. No correlation was found for the septal and lateral walls. NAC, non-attenuation-corrected.

In 13% of cases, the nonrigid correction increased the anterior wall intensity by 10% or greater. Figure 6 shows representative SA slices for a case in which respiratory motion correction increased the anterior wall intensity by 26% for the ACGCT images and by 28% for the ACpseudo images.

Fig. 6

Short-axis slices for the (a) nonregistered ACGCT and (b) registered ACRegGCT PET images. Motion correction increased the anterior and inferior wall maximum by 26 and 19%, respectively. AC, attenuation correction.

When compared with the ACHCT images, an anterior wall intensity increase of greater than 10% was seen in 14% and 23 of cases for the ACRegGCT and ACRegpseudo images, respectively.

Discussion

PET-CT misalignment can lead to AC errors, which in cardiac imaging can present as false defects. Compared with the HCT, the ACT, GCT and pseudo-gated CTs, on average, improved the liver and cardiac PET-CT alignment. Importantly, however, all methods were also shown to worsen the alignment in some cases. As a consequence of worse alignment, the ACT decreased the anterior wall intensity in AC PET images by more than 10% in 21 out of 60 cases, which is comparable to findings from other groups; Alessio et al. 15 found that in 23% of cases the ACT produced unacceptable PET-CT alignment, and Gould et al. 5 found artefactual defects in both HCT and ACT data sets in 19% of cases.

A high correlation between the change in cardiac PET-CT alignment and the change in wall intensity was found for the anterior and lateral walls. PET-CT misalignment of these walls is expected to locally reduce the intensity in AC PET images, as undercorrection of attenuation results from replacing myocardium μ values with smaller lung μ values. No cases were found that suggested overcorrection of the lateral and anterior walls, in which a superior displacement of the myocardium would increase anterior or lateral wall intensity. We can therefore assume that an increase in the intensity of the lateral and anterior walls reflects an improvement in accuracy, and a decrease indicates a loss of accuracy.

A significant decrease in inferior wall intensity was found for the ACACT and ACGCT data; however, a small but significant increase was found for the ACpseudo data. Misalignment of the liver is known to affect the inferior wall, and the pseudo-gated CT produced the best liver alignment. All three cine CTs demonstrated cases in which liver alignment worsened, by up to 13.1 mm for the ACT and GCTs; however, for the pseudo-gated CT the largest degradation of alignment was only 1.6 mm. Additional to liver alignment, we also observed that PET-CT misalignment can result in incorrect AC of the inferior wall with fat, and this could contribute to the reduction in inferior wall activity seen in ACACT and ACGCT images with poorer PET-CT alignment.

A degradation of alignment and resulting decrease in anterior wall intensity occurred in more cases and to a greater extent when the HCT was replaced with the ACT and GCT, compared with the pseudo-gated CT, which produced fewer and less severe cases. Conversely, when the alignment improved and the intensity of the anterior wall increased as a result, the ACT and GCT had a greater effect than the pseudo-gated CT. This is reflected in the 21 cases in which the ACACT PET images produced more than 10% increase in anterior wall intensity when compared with ACpseudo.

Of the AC methods evaluated in this study, the pseudo-gated CT was found to be the most robust. Regardless of the AC approach used, visual assessment of PET-CT alignment appears necessary as misalignment can still exist. In addition, images that demonstrated voluntary motion were excluded from this study, further highlighting this requirement.

Despite the significant correlation between the effect of respiratory motion correction and the range of respiratory-induced displacement, the cases in which respiratory motion correction produced the largest change in inferior and anterior wall intensity were not necessarily the cases with the largest motion range. A likely explanation for this is the variation in respiratory characteristics between patients, as both the amplitude of motion and the amount of acquisition time spent at maximum displacement will determine the effect of motion in the PET image. Liu et al. 40 found that PET images of patients who spent a large portion of the respiratory cycle at expiration phase were less affected by respiratory motion than were images of patients with irregular breathing patterns. In our study, respiratory motion correction did generally produce a smaller average effect when a large portion of time was spent at expiration; however, exceptions to this were found. Additional factors could explain this, including differences in myocardial wall thickness and the residual blurring in PET gates. Moreover, a variable amplitude-gating method was used that produces gates of equal noise, and a potential consequence of this is increased motion blurring in some gates.

The findings of this study are confined to the specific methods of respiratory motion reduction used. Simple approaches to each concept were implemented, and improvements to each of the methods may alter the results. Average cine CT was chosen, as this is the method currently available on the scanner used in this work; however, a maximum intensity projection cine CT has been shown to produce superior results 15. Cine CT images were gated with a data-driven gating method, and the ACGCT PET images were similar to the ACACT images. An improvement to the gating could improve PET-CT misalignment, and hardware gating methods or other data-driven CT gating methods may produce superior results 41–45. Finally, the pseudo-gated CT was derived in a simple manner by applying PET-derived motion fields to the HCT. McQuaid et al. 28 have produced a detailed method of generating cine CTs from static CT images, and Alessio et al. 46 have shown that consistency criteria can also be used to reduce respiratory motion artefacts in cardiac PET imaging. These approaches may improve the results for the pseudo-gated CT approach.

Limitations of the methods used for measuring the PET-CT misalignment and measuring cardiac wall profiles must also be considered. Measuring the edge of an object in a PET image is not well defined because of resolution effects, and in this work the edge was approximated as the midpoint between the local minimum and maximum on either side of the edge. This could be a source of error in the PET-CT misalignment values; however, the results demonstrate that, generally, the relative change in cardiac wall measurement approached zero as the PET-CT misalignment approached zero, supporting the accuracy of the approximation. Alignment was, however, only calculated at one position on the myocardial wall and liver, which may not truly represent the alignment throughout the entire area of interest, and cases were found in which the alignment at the basal region did not represent alignment at the apical region. Similarly, only eight cardiac wall profiles were measured, and some changes in the myocardium may have been undetected.

Respiratory motion correction did not degrade image quality (by reducing cardiac wall intensity by >10%), and when correcting the ACpseudo and ACGCT PET images for respiratory motion 13% of anterior wall profiles increased by more than 10%. This seems low; however, only the effect of blurring has been considered. On correcting for both PET-CT mismatch and blurring (by comparing ACRegpseudo to ACHCT), we found that the anterior wall intensity increased by more than 10% in 23% of cases.

This implementation of respiratory motion correction could potentially improve; gated and reconstructed images were individually registered to the expiration gate and summed to create the motion-corrected image. Other registration methods such as optical flow 21, combining the registration within iterative reconstruction 22–24,47,48 and combining the generation of a pseudo-gated CT with registration and reconstruction 49 all potentially improve the results.

Conclusion

In general, more accurate AC by improving cardiac PET-CT alignment increased the anterior and lateral cardiac wall intensity measured on AC PET, and worse PET-CT alignment, resulting in less accurate AC, decreased the intensity of these walls. The inferior and septal walls were prone to both overcorrection and undercorrection for attenuation. Compared with the standard approach of using an HCT for AC, on average, all three methods of reducing respiratory motion-induced AC errors improved PET-CT alignment and increased the anterior wall intensity. However, all three methods also produced cases in which alignment worsened, and in some cases a severe artefact appearing as a defect resulted. This occurred in more cases and to a greater extent for the ACT and GCT methods, and the pseudo-gated cine CT was found to be the most robust method for reducing respiratory-induced AC artefacts in NH3 cardiac PET-CT imaging.

Respiratory motion correction generally increased the inferior and anterior wall intensity, and the amount of change was correlated with the range of respiratory motion. Although respiratory motion correction produced a large effect only in a small number of cases, up to 41% change in anterior wall intensity was demonstrated. Furthermore, no degrading effects were found when respiratory motion correction was applied.