In the first part of this study, we investigated the ability to visually assess and manually correct patient motion in 160 dynamic 15O-water PET scans with simulated motion mixed with 10 motion-free scans. All data were analyzed using two different motion correction approaches. Motion had a significant effect on MBF, as illustrated by the heat maps in Figure 2. The patterns of MBF deviations with negative median relative deviations in RCA and the largest deviations in LAD agreed with the previous study by Nordström et al.6

The post-analysis approach led to fewer corrections

The motion correction problem consists of two parts: (1) Identifying significant motion and (2) correcting the motion. In the simulation study, despite a high prevalence of motion (94%), only 63 scans (37%) exhibited significant motion artifacts on the coronary territory level before correction. When using the pre-analysis approach, without any knowledge of the pre-correction results, nearly all scans (94%) were motion-corrected. However, when the post-analysis approach was used, in which the pre-correction results were reviewed, the number of corrected scans was only 64%. Many scans with motion were not corrected using the post-analysis approach, implying that scans without perceived motion artifacts in parametric images were left uncorrected.

Large corrections performed mainly in the axial direction

The simulated data included motion isolated to the y-direction in 35%, to the z-direction in 35%, and in both the y- and z-direction in 24% of the scans and no motion in the x-direction. Yet, when using the pre-analysis approach, corrections were applied in the x-direction in 32% of the scans (in 8% > 10 mm), suggesting that visually identifying the direction of motion was difficult. When using the post-analysis approach, correction in the x-direction was applied to only one scan, however, this may be due to observer variation rather than the approach. Despite an equal distribution of simulated motion between the y- and z-directions, corrections were mainly applied in the z-direction for both approaches (44% in y vs 89% in z and 28% in y vs 44% in z, respectively). Similarly, large corrections (> 10 mm) were mainly applied in the z-direction for both approaches (25% in y vs 65% in z and 14% in y vs 28% in z, respectively), indicating that motion in the z-direction was more easily detectable. Both image readers were generally too conservative when it came to the magnitude of corrections and residual motion was found in most scans after correction (Figure 3). Both approaches tended to undercorrect more types of motion in the y- and z-directions and overcorrect fewer types of motion. The undercorrection was observed for the same types of motion regardless of the approach used.

Large artifacts were reduced

For large motion artifacts, both approaches yielded similar results as artifacts in 84% and 82% of the scans were reduced, respectively. Out of six scans with very large artifacts (> 80%), the motion artifact was reduced in five for both approaches. The reduction of motion-induced artifacts does not imply complete elimination of the artifact. In fact, the median reduction was only 46%, implying significant residual artifact in most scans. An example of successful correction of a large motion artifact is shown in Figure 4.

Figure 4

Polar plots from a patient scan comparing images pre-correction (A), post-correction (B), and original motion-free (C). The simulated linear slide motion (20 mm, 1 minute post-injection) increased MBF in the anterior part and reduced MBF in the inferior part with a large apical inferolateral defect (A). After motion correction, the artifact disappeared (B), aligning with the original motion-free image (C)

Pattern of deviation in RCA and LAD persisted after correction

The impact of motion correction on median MBF deviations is visualized in Figure 2, showing a reduction in deviations after correction (Figure 2A–C) as expected. However, significant deviations in MBF persisted in most motion types, and the pattern of negative deviations in RCA and positive deviations in LAD remained. As for the large artifacts, motion correction could only reduce and not completely remove motion artifacts. Significant MBF deviations introduced by the motion correction were induced in three cells using the pre-analysis approach (Figure 2H). Two of the new deviations were in LAD and all in scans with motion in the anterior (+ y) and/or cranial direction (+ z). They were, however, negligible with median deviations below 5%. Even though no new significant deviations were introduced by using the post-analysis approach, it proved to be slightly less consistent in reducing or removing artifacts in RCA, caused by discomfort to the stress agent (Figure 2I) compared to the pre-analysis approach. This suggest that these artifacts were difficult to identify in some of the post-analysis parametric images. Nevertheless, median and maximum deviations were similar for both approaches in the RCA.

Care should be taken when using hard cut-offs in cases with motion

Based on a MBF cut-off at 2.3 mL⋅min−1⋅g−1, 60% of the original motion-free scans were positive for reduced MBFstress globally or in one or more coronary territory. This is higher than to be expected in a clinical cohort.2 In addition, median MBF of the original motion-free scans globally and in all territories were close to 2.3 mL⋅min−1⋅g−1. As an example, one original scan had a global MBF of 2.31 mL⋅min−1⋅g−1, and since it was represented 17 times, it resulted in 17 scans that could potentially become positive by a small reduction in MBF of only 0.02 mL⋅min−1⋅g−1. As a consequence and despite generally reduced motion artifacts, when applying either the pre- and post-analysis approach, half of the scans that underwent interpretation changes following motion correction in this study were incorrectly reclassified (Table 2). However, while the number of changes in interpretation may seem disconcerting, the absolute differences in MBF values were, as mentioned above, small.

In future studies, it may be useful to investigate the impact of patient motion on MBF in both high-and low-risk populations to better understand the potential impact of patient motion on clinical decision making.

Post-analysis approach is the most feasible implementation

Overall, the impact of the pre- and post-analysis approaches were similar and large artifacts were identified and reduced to a similar extend. However, the post-analysis approach resulted in a large reduction in number of scans with applied correction. By adopting this approach, the potential benefits of motion correction can be achieved without unnecessarily intervening in motion-free scans, making it the most feasible implementation in a clinical setting.

Reported number of motion cases in the clinical cohort was low

In the second part of the study, the clinical effect of motion correction was assessed in a retrospective review of 1545 clinical patient exams. Fifty exams (3%) were reported as ‘conclusive with reservations due to motion’ and only 14 (1%) exams were categorized as ‘inconclusive due to motion.’ Combining these two categories, the prevalence of motion in the current study was 4%, which is relatively low, as several groups reported a prevalence of 30-68%.6,7,8,9,10,17 Multiple factors play into these discrepancies. The current study was retrospective, and readers were not instructed to report all motion. Instead, motion was only reported if it was expected to have consequences for the reading and small artifacts or artifacts that would not influence the interpretation of the exams were ignored. This contrasts with studies where specific tests for motion were performed on all scans resulting in a much higher prevalence.

Interestingly, when the motion correction tool was provided, 80 out of the 762 exams were motion corrected, corresponding to a prevalence of 10%. This points to motion being more prevalent than indicated in the clinical exam reports, but also suggests that the readers were not certain about the significance of motion. Of the 80 clinical patient scans where motion correction was performed, MBFstress below the threshold of normal myocardial perfusion of 2.3 mL⋅min−1⋅g−1 before motion correction in 44% on the global level and in 59% on the territorial level. In scans with reduced MBFstress in only one coronary territory, the reduction was predominantly in RCA, which aligns with previous findings from simulation studies and other research that have shown that motion affects MBF particularly in RCA.7,8,17

Motion correction in the clinical cohort had little impact on interpretation

A clinical patient exam is displayed in Figure 5, illustrating the commonly observed motion artifact pattern with apparent anterior hyperperfusion and wall thickening combined with inferior hypoperfusion and wall thinning, found in 16% of exams. As shown in the simulation study, a range of stress agent responses and linear slide motions give rise to this artifact pattern. Pre-correction, the RCA defect was 20% of the LV, but after motion correction, it was eliminated. In most of the clinical cases (83.8%), motion was identified and corrected in the caudal (− z) or cranial (+ z) direction. This aligns with previous research reporting a frequent occurrence of patient motion in the z-direction,9,23 but may also reflect that motion in the z-direction is more easily identified compared to x and y motion as seen in the simulation study (Figure 6).

Figure 5

PET images representing a patient from the clinical cohort pre- (A, B) and post-motion (C, D) correction. In the polar plot in A, a motion artifact has caused a false positive defect (20.1 %) in the RCA territory. In the post-motion correction polar plot in C, the motion artifact is completely reduced (defect 0.0 %). Splash images of short axis, horizontal long axis, and vertical long axis in B demonstrate the effect of motion artifacts on the visual interpretation of the images. The inferior wall appears hypoperfused compared to the rest of the myocardium. The post-motion correction splash images in D, demonstrate a uniform tracer uptake, thereby eliminating any suspicions of defects in the inferior wall. MBF, myocardial blood flow (mL⋅min−1⋅g−1); PTF, perfusable tissue fraction (mL⋅mL−1)

Figure 6

Scatter plots and Bland–Altman plots of regional MBFstress as well as global MBFstress for precorrection vs motion-corrected images in the clinical patient cohort. Dashed lines in the scatter plots indicate the threshold of normal values (MBF ≥ 2.3 mL⋅min−1⋅g−1). Red lines represent lines of identity. Dashed lines in the Bland–Altman plots represent limits of agreement

Motion correction had minimal impact on the interpretation of most exams. Only five of 240 coronary territories changed from positive (< 2.3 mL⋅min−1⋅g−1) to negative (≥ 2.3 mL⋅min−1⋅g−1) and eight changed from negative to positive after motion correction. However, some changes were in the same exam and, in only six exams, readings changed from hypoperfusion in one or more coronary territories to normal perfusion in all territories (n = 4) or vice a versa (n = 2). An additional four exams had an increase in the number of territories with hypoperfusion corresponding to a total impact of the correction in 12.5% of the motion corrected exams or 1% of the entire clinical patient cohort.