We evaluated the use of a novel 3 × MC PET reconstruction technique tested for the first time in PET imaging of aortic valves, including cardiorespiratory and gross patient motion correction. The impact of the 3 × MC protocol was evaluated on five criteria assessing the quantitative and qualitative assessments of the images: SUVmax, TBRmax, SNR, the test–retest repeatability, and the correlation between the quantitative measures and AVCS. The 3 × MC was superior to the standard (end-diastolic) and ECG-MC imaging protocols without affecting the test–retest repeatability. The improved localized uptake observed in the aortic valves for the 3 × MC reconstructions correlated better with the AVCS than the standard and ECG-MC protocols, which in combination with the increased SNR might aid the understanding of the pathophysiology in native and bioprosthetic valve diseases.
The assessment of aortic valve microcalcification with 18F-NaF can be used to predict aortic valve stenosis progression [1, 4, 16, 21]. Studies to date have established the association between increased TBRmax and aortic stenosis progression using either standard (end-diastolic) or ECG-MC image series [1, 10, 15]. However, it is uncertain how much patient and respiratory motion can affect quantitative assessments of 18F-NaF uptake (SUVmax and TBRmax). It is therefore of great interest to evaluate potential image improvements offered by the 3 × MC protocol in the context of aortic valve microcalcification. Specifically, the improved co-localization of PET activity with areas of aortic valve calcification is of high interest in studies investigating the various causes of leaflet calcification, i.e., mechanical stress at the leaflet coaptation points versus commissures. Moreover, it might help differentiate between activities originating in the aortic valve from activity from nearby structures (left main stem, mitral valve, left atrium, etc.) by reducing the spillover effect of those structures. This is of importance in cases of bioprosthetic valves, and transcatheter aortic valve implantation (TAVI), where localizing the source of PET activity can help differentiate between bioprosthetic valve leaflet degeneration and remote activity originating from the valve struts or crushed native leaflets in case of TAVI [22]. In this context, partial volume effects strongly affect the activity observed in the aortic valves and their surrounding tissues. It is anticipated that the 3 × MC reconstruction protocol may ameliorate the impact of the apparent partial volume effects because of the co-registration of the gated images with significantly reduced intra-gate motion compared to the standard and ECG-MC reconstruction protocols where several motion patterns blur the resulting images. Therefore, it is believed that the 3 × MC reduces the impact of the partial volume effects while also aiding toward better test–retest repeatability.
In the current study, all patients demonstrated an increase in both SUVmax and TBRmax when the data were corrected using 3 × MC (cohort-based increase: SUVmax = 26%, TBRmax = 33%) (Fig. 3), indicating loss of signal when not applying these correction techniques. The improved SUVmax and TBRmax assessments for the 3 × MC both had preserved repeatability measures and led to a twofold increase in the correlations to the aortic valve calcium score when compared to the Standard and ECG-MC reported results; thus, suggesting that 3 × MC might improve predictions on disease progression (Figs. 3, 4, 6, 7, 8) [1]. Based on these findings, we can conclude that the motion patterns across different patients vary widely, depending on the respiratory translations and patient motion patterns during the acquisitions, which in some cases can introduce variations in the quantitative values exceeding 40% (average increases of the four scans with substantial changes in SUVmax and TBRmax values: SUVmax = 42%, TBRmax = 72%). The increase in the correlation scores observed for the TBRmax and aortic valve calcium score may translate into lower number of patients required in studies investigating the effects of interventions on 18F-NaF PET uptake (as a marker of calcification activity) because any true effect will not be covered by noise within the region of interest.
Motion during the scans has been shown to have a detrimental impact on the image quality, measured as SNR [10, 13]. In concordance with a previous study of coronary plaques, the SNR was significantly reduced for the standard imaging protocol when compared to the motion-corrected protocols [13]. The low SNR observed for the standard protocols is introduced by two arms, low count statistics (25% of the acquired data used for the analyses), and respiratory and patient motion blur embedded in the ECG-gated images [12]. Introducing motion correction (ECG-MC and 3 × MC), the SNR improved as all the data were used in the analyses partly owing to the increased count statistics in the images (100% of the acquired data). The further improvement observed for the 3 × MC was introduced by the corrections for both respiratory and patient motion, which reduced the residual blur introduced to the ECG-gated reconstructions, as shown in Fig. 4. While SNR was significantly reduced using ECG-MC alone compared to 3 × MC, ECG-MC provides greatly improved SNR when compared to the standard assessment and therefore should be considered when 3 × MC is not possible. Of note, the gated reconstructions took 3 min to reconstruct each gate using a 3 year-old PC, introducing a reconstruction time of 48 min per repositioning event.
While some of the discrepancies, in general, may be attributed to changes in the pathological disease, only minor changes in the SUVmax/TBRmax test–retest variability is expected to be introduced by disease progression owing to the slow microcalcification processes which may change the calcification burden by 24% per year (and thus, only 1–2% change in the SUVmax/TBRmax burden should expected) [16]. Therefore, the discrepancy in the impact of the ECG-MC and 3 × MC observed for SNR, TBRmax, and SUVmax indicate that the different motion patterns (cardiac contraction, respiratory and gross patient motion) affect image quality to varying extent. The improved quantitative and qualitative findings observed for the 3 × MC protocol suggest that cardiac contraction is of less importance to correct for in studies of aortic valves when compared to the patient repositioning events and the respiratory translations. Such results were expected as the cardiac contraction affects mainly the coronary arteries, where the right coronary artery has been reported to shift up to 26 mm [23] which is in contrast to the average displacement of the aortic valves (~ 12 mm) [24]. In contrast, respiratory motion has been reported to displace the heart by up to 21 mm [25] and patient repositioning events might shift the heart 5–15 mm [11]. In this context, patient motion (repositioning events) usually happens with low frequency (approximately 3.5 times per scan) and although the translations are few, they introduce non-periodical shifts of the respiratory baseline and, thus, affect the cardiorespiratory motion correction.
Another important finding in our study is the robustness of the data-driven motion detection techniques, as applied to valve imaging studies. A previous study has shown that motion detection techniques employing these assessments are affected by reductions in the count rates [26]. The count rates are mainly affected by two variables: the injected activity and injection-to-scan delay. While the previous study utilizing a 3 × MC protocol (focusing on coronary plaques) had similar injection-to-scan delays as the current valve study (59–99 min) [12], the injection doses, in general, are halved for the studies of aortic valve microcalcification (to 125 MBq) [7, 10, 17], as compared to studies of coronary plaques (250 MBq) [12, 14, 27]).
Given the large dose-reduction in the current study, it was important to test the robustness of the extracted motion patterns (respiratory and patient). In this context, we evaluated the robustness of the technique by evaluating the impact of the data-driven motion detection techniques through assessments of the TBRmax, SNR, and test–retest repeatability of the TBRmax assessments. First, the number of repositioning events reported in this study was in concordance with previous studies [11, 12], suggesting that the robustness of the motion detection is preserved. Motion correction of the detected repositioning and respiratory events introduced an increase in both SUVmax, TBRmax, and SNR while preserving a high test–retest repeatability which is in concordance with already established imaging protocols [28]. These findings strongly indicate that the data-driven motion detection technique developed for patient and respiratory motion detection algorithm provides robust and reliable results in valve imaging even when using a low-dose imaging protocol.
Limitations
In this study, the number of patients was limited to 14 who underwent two PET/CTA scans within a month—this number is limited by the difficulty in obtaining such repeated PET/CT scans with short time interval. Nevertheless, we were able to report substantial improvements for SUVmax and TBRmax following 3 × MC, which correlated better to the calcium scores. The attenuation correction was not motion-corrected prior to image reconstruction, which might pose another limitation. However, in a previous study from our center, we showed that a respiratory averaged CT attenuation correction did not change the quantitative assessment. Therefore, we do consider this a limitation of this study [29]. Another limitation was the use of only four cardiac and four respiratory gates for each patient repositioning event during the scans; however, double gating imposes count limitations especially in low-dose studies as studied. Further improvement in TBRmax values is possible with an increasing number of cardiorespiratory gates or introducing motion correction during the reconstruction, whereby noise in the images will be suppressed.