Evaluation of the second-generation whole-heart motion correction algorithm (SSF2) used to demonstrate the aortic annulus on cardiac CT

The main purpose of pre-transcatheter aortic valve implantation (TAVI) cardiac computed tomography (CT) for patients with severe aortic stenosis is aortic annulus measurements. However, motion artifacts present a technical challenge because they can reduce the measurement accuracy of the aortic annulus. Therefore, we applied the recently developed second-generation whole-heart motion correction algorithm (SnapShot Freeze 2.0, SSF2) to pre-TAVI cardiac CT and investigated its clinical utility by stratified analysis of the patient's heart rate during scanning. We found that SSF2 reconstruction significantly reduced aortic annulus motion artifacts and improved the image quality and measurement accuracy compared to standard reconstruction, especially in patients with high heart rate or a 40% R-R interval (systolic phase). SSF2 may contribute to improving the measurement accuracy of the aortic annulus.

www.nature.com/scientificreports/ multi-segment reconstruction; software solutions have been developed 11 . A recent study reported that a generative adversarial network model could create images with fewer motion artifacts while preserving the lesion contrast 12 . Although such an approach using machine learning may effectively reduce cardiac motion artifacts, it has not yet been implemented in clinical practice. The first-generation motion correction algorithm (SnapShot Freeze, SSF1; GE Healthcare) is vendor-specific and designed to address coronary motion artifacts on cardiac scans. Its application significantly improved the image quality of the coronary arteries in patients with a high heart rate [13][14][15][16][17][18][19][20] . However, SSF1 cannot address other non-coronary intracardiac structures. The second-generation motion correction algorithm (SnapShot Freeze 2.0, SSF2; GE Healthcare) increased the motion-correction range to include the whole heart within one scan volume 21,22 .
To our knowledge, the clinical effectiveness of SSF2 with respect to the aortic annulus has not been investigated. We hypothesized that use of the SSF2 algorithm would improve the image quality of CT scans acquired to evaluate aortic valves, because a reduction in motion artifacts can improve the CT image quality and diagnostic accuracy. In this study we enrolled patients with severe aortic stenosis and compared the quality of standard images without motion correction with pre-TAVI cardiac CT scans subjected to SSF2.

Materials and methods
This retrospective study conformed to the principles of the Declaration of Helsinki. In accordance with the Ethical Guidelines for Medical and Health Research Involving Human Subjects (Ministry of Education, Culture, Sports, Science and Technology and Ministry of Health, Labour and Welfare, Japan), study information including the objectives was disclosed on our hospital website with an opt-out approach. The Ethical Committee for Epidemiology of Hiroshima University reviewed and approved the study protocol (No. E-2623, Clinical study of motion correction algorithm for cardiac CT). Informed patient consent for the analyses was waived.
Study population. We enrolled 108 patients with severe aortic stenosis who underwent cardiac CT as candidates for TAVI between April 2021 and February 2022. Inclusion criteria were patients who underwent contrast-enhanced cardiac CT. Our exclusion criteria were severe renal failure (estimated glomerular filtration rate < 30 ml/min/1.73 m 2 , 15 patients), poor breath holding during scanning (1 patient), extravasation during contrast injection (1 patient) or refusal of CT examination (1 patient). Thus, the final study population consisted of 90 patients; they were 33 male and 57 female ranging in age from 70 to 95 years (median age, 84 years).
CT scanning. All patients were scanned on a 256-detector row CT scanner (Revolution CT; GE Healthcare, Milwaukee, WI, USA); prospective electrocardiogram-gated axial scans were acquired. As shown in Table 1, the scanning-and reconstruction parameters were tube voltage, 120 kVp; tube current, selected by automatic tube current modulation (Smart-mA, GE Healthcare) based on the scout image; noise index, 25; detector collimation, 256 × 0.625 mm or 224 × 0.625 mm depending on the patient's heart size; gantry rotation, 0.28 s; slice thickness, 0.625 mm; scan field of view, 360 mm; display field of view, 200 mm; matrix, 512 × 512; reconstruction, half; reconstruction kernel, HD standard; reconstruction method, deep learning image reconstruction (TrueFidelity, strength High; GE Healthcare) [26][27][28][29] . The padding range was 0-100% of the R-R interval when a heart rate of less than 60 bpm was recorded during pre-examination monitoring; when it exceeded 60 bpm or was variable. In the presence of arrhythmia the padding range was 0-250%. All scans were craniocaudal from the tracheal bifurcation to the level of the inferior margin of the cardiac apex. All patients were able to perform breath-holds during the examination. To achieve high image quality with minimal radiation doses, patients with a heart rate above 60 bpm 5 min before the start of scanning were given 2-10 mg propranolol hydrochloride (Inderal; Taiyo Holdings Co., Ltd., Tokyo, Japan). www.nature.com/scientificreports/ The contrast medium (iodine concentration 350 mg/ml; Iomeron-350; Eisai Co., Ltd., Tokyo, Japan) was injected in triple-phase through a 20-or 22-gauge catheter into the antecubital vein using a power injector (Dual Shot type GX; Nemoto Kyorindo, Tokyo, Japan). The iodine dose of 273 mg/kg in the first phase was administered in 13 s. This injection was followed at a speed of 5 s by a 50/50 mix of contrast medium (53 mgI/kg) and saline, and finally 100% saline was delivered at the same injection speed. For cardiac CT scanning, bolus-tracking was used to synchronize the arrival of contrast medium at the left atrium and left ventricle with the start of scanning. To monitor arrival, we acquired axial scans at one-second intervals at the level of the left atrium and the left ventricle 10 s after the start of contrast medium injection. The radiation dose was 120 kVp, 40 mA. Scanning was started automatically 5 s after contrast enhancement reached 150 Hounsfield units in a region of interest (ROI) placed in the center of the left atrium and the left ventricle.
Data processing. Similar to the SSF1 algorithm 18,20 , the SSF2 algorithm uses data from adjacent cardiac phases (64 ms before and after the target phase) to characterize and correct the motion. The SSF2 algorithm, a fully automated technique based on information and feedback obtained from SSF1 scans, seeks each region at all image volumes for a local path that is consistent with the subset of measured data. Once the vessel's motion path is identified, the data are discretized into a series of datasets based on when the corresponding projection rays were measured. Each volume dataset in the series undergoes spatial deformation by the motion field. This allows the motion state to be mapped from the respective time to the central reference time, which is determined by the prescribed cardiac phase 30 .
All images were reconstructed using the standard (without motion correction) algorithm with deep-learning image reconstruction for reducing the image noise [26][27][28][29] . For the cardiac phase, the systolic-(R-R interval, 40%) and diastolic phase (R-R interval, 75%) used for pre-TAVI cardiac CT measurements were selected 2,4,30,31 . As the systolic-and diastolic phases were additionally subjected to SSF2 reconstruction, 4 datasets were obtained for each patient. They were anonymized and transferred to the workstation (Advantage Workstation 4.7, GE Healthcare) for later analysis.
Quantitative evaluation. The attenuation effect elicited by motion artifacts was analyzed at the aortic annulus. All images were inspected by one radiological technologist (Y.M. with 15 years of experience with cardiac CT studies). To assess the aortic annulus, only axial-and 2D double-oblique multiplanar reconstruction (MPR) images were examined. The aortic annulus was defined as a virtual ring formed by joining the basal attachments of the aortic leaflets 2,32 .
Edge rise distance. We generated a 3-directional CT attenuation profile (anterior-, superior-, and inferior direction) of the aortic annulus ( Fig. 1) using the particle analysis tool (Plot Profile) on the workstation (Ziosta-tion2, Ziosoft, Tokyo, Japan). Areas of calcification where CT attenuation fluctuates significantly were carefully www.nature.com/scientificreports/ avoided. The CT attenuation profiles were generated at precisely the same location for images reconstructed with standard and SSF2. We cut off the bottom and top 10% of the profile and measured the 10-90% edge rise distance (ERD) 33,34 . The ERD was examined in three directions of the aortic annulus and the mean values were compared on standard-and SFF2 images.

Dispersion of sizing.
With respect to the sizing of the aortic annulus, we evaluated the dispersion between the two reconstructions. All images were analyzed by two radiological technologists (Y.M. and C.F., with 15 and 18 years of experience in cardiac CT imaging, respectively). They were blinded to presence of SSF2 technique and manually measured the aortic annulus area and perimeter of all patients independently on a CT workstation (Ziostation2, Ziosoft, Tokyo, Japan). The annulus area and perimeter were determined from the trace by placing plots (dots) at the blood-tissue interface 1,9 . The annulus border was traced outside calcifications. When boundaries were unclear due to motion artifacts the observers traced subjectively. After the plots and annular borderline were drawn manually, the annulus area and perimeter were automatically calculated on the workstation 1,9 . Contrast-to-noise ratio. To investigate the potential effect of SSF2 reconstruction on the quantitative ERD measurements, we inspected axial images and recorded the CT number and image noise [standard deviation (SD) of the CT number] in a circular ROI placed in the ascending aorta and septal wall of the ventricle. The size of the circular ROI cursor was as large as allowed by the diameter of the ascending aorta (approximately 5.0-10.0 mm 2 ) and of the septal wall of the ventricle (approximately 1.5-3.0 mm 2 ). Based on the obtained values we also calculated the contrast-to-noise ratio (CNR) using the formula: (CT number of the ascending aorta minus the CT number of the septal wall of the ventricle) divided by the image noise of the ascending aorta 35 .
Qualitative analysis. Two radiological technologists (Y.M. and C.F., with 15 and 18 years of experience in cardiac CT imaging, respectively) were blinded to presence of the SSF2 technique. They subjectively and independently inspected the MPR images from the sinotubular junction to the left ventricular outflow tract of the datasets for motion artifacts at the aortic annulus level. To grade the image quality they used the 5-point Likert scale where 1 = very poor (motion artifacts resulting in poor visualization of the aortic valve anatomy, not evaluable), 2 = poor (degraded visualization of the aortic valve anatomy due to motion artifacts, not evaluable), 3 = fair (minor motion artifacts with clear delineation of the aortic valve anatomy), 4 = good (no motion artifacts with confident identification of the aortic root anatomy including the cusp nadirs and annular contours), and 5 = excellent (outstanding image quality with a high level of diagnostic certainty with regard to the aortic valve cusps, the leaflet nadirs, and the detection of the aortic annular contours) 30 . Interobserver disagreement was resolved by consensus.
Statistical analysis. Continuous variables of demographic data, ERD, CT number, image noise and CNR are expressed as the median and range or as percentages or counts, aortic annulus area and perimeter or image quality scores as the mean and SD. The results of ERD, CT number, image noise, CNR and image quality scores were compared on images reconstructed with standard and SSF2 using the Mann-Whitney U-test. To compare the dispersion (SD) of area and perimeter between the two reconstructions we used the F-test. Intra-and interobserver agreement on the variability of the quantitative evaluation by Bland-Altman plot analysis was expected to converge to a 95% limit of agreement, defined as a mean difference of ± 1.96 SD. To determine whether the CNR was equivalent in standard and SSF2 reconstructions, we performed the equivalence test 36 . As the SD of the CNR between the proximal coronary arteries and the adjacent perivascular tissue was 5 in our earlier study 34 , we adopted 5 as the equivalent margin. Interobserver agreement in the qualitative evaluation was classified as evaluable (score 3-5) and non-evaluable (score 1, 2) assessed with the Cohen kappa κ coefficient where a κ value of less than 0.20 = poor, 0.21-0.40 = fair, 0.41-0.60 = moderate, 0.61-0.80 = substantial, and 0.81-1.00 = near perfect agreement. All statistical analyses were performed with JMP 16 (SAS Institute Inc., Cary, NC, USA). Differences of p < 0.05 were considered statistically significant.  Table 3. In patients with a low heart rate, the ERD obtained with standard and SSF2 reconstruction was not significantly different (R-R 40% and R-R 75%: p > 0.05). However, in patients with an intermediate heart rate, the ERD at R-R 40% was significantly shorter on SSF2 (2.0 mm)-than standard (2.4 mm) images (p < 0.001). In patients whose heart rate was high, the ERD at R-R 40% and R-R 75% was significantly shorter on SSF2-than standard images (p < 0.001). Bland-Altman plots for intraobserver agreement with respect to the ERD for 4 datasets are summarized in Fig. 2. The plots nearly converged within the 95% limit of agreement for all datasets.  Table 5, the SD of the aortic annulus perimeter was also significantly smaller in SSF2 reconstruction than in For the area and perimeter, the plots almost converged within the 95% limit of agreement for all datasets. In particular, the mean difference (± 1.96 SD) was within a smaller range on SSF2-than standard reconstruction images.

Dispersion of sizing. As shown in
Contrast-to-noise ratio. As shown in Table 6, the CT number of the ascending aorta and the septal wall of the ventricle and the image noise of the ascending aorta showed no significant difference between the two reconstructions, irrespective of the patients' heart rate (all: p > 0.05). In addition, these CNR also showed no significant difference between the two reconstructions at low ( 7 to 0.9 in patients with a high heart rate. Because the 95% confidence interval did not cross the bilateral predefined equivalence margin (Fig. 5) in all heart rate classes, we considered CNR to be equivalent among our standard and SSF2 reconstitution irrespective of their heart rate.
Qualitative analysis. Table 7 shows the results of the visual evaluation of MPR images submitted by our two readers. In patients with a low heart rate, at R-R 75%, there was no significant difference in the mean image scores assigned to images subjected to standard-or SSF2 reconstruction (p = 0.540). At R-R 40% the visualization scores were significantly higher for images reconstructed with SSF2 than standard (all: p < 0.01). There was substantial interobserver agreement with respect to the overall image quality (κ = 0.69). SSF2 reconstruction improved the image quality of the aortic annulus in the representative case shown in Fig. 6. All results are provided in the Supplementary file.

Discussion
Our study demonstrates that the second-generation whole-heart motion correction algorithm (SSF2) was superior to standard reconstruction with respect to the image quality of pre-TAVI cardiac CT scans acquired for the evaluation of the aortic annulus.

R-R interval (%) Standard SSF2 P
Patients with low HR (< 60 bpm)  A and B) and standard-and SSF2 reconstruction at an R-R interval of 75% (C and D). The solid line represents the mean difference, the dashed lines represent the 95% limit of agreement (mean difference ± 1.96 SD). SD = standard deviation. www.nature.com/scientificreports/ At R-R 40%, SSF2 reconstructed images received significantly higher image quality scores than did standard reconstruction regardless of the patients' heart rate (p < 0.001). At R-R 75%, in patients with an intermediate and high heart rate was the visualization score higher for SSF2-than standard reconstructed images. At R-R 40% and R-R 75%, SSF2 strongly tended to yield higher image quality scores than did standard reconstruction. Consequently, SSF2 reconstruction raised the image quality significantly, especially in patients with a high heart rate or a 40% R-R interval.
The earlier vendor-specific motion correction algorithm (SSF1) was designed to address coronary motion artifacts on cardiac scans. It was primarily indicated for coronary imaging and was shown to improve the image quality and diagnostic accuracy of scans performed for the detection of significant coronary stenosis, especially in patients with a high heart rate [13][14][15][16][17][18][19][20] . The SSF2 algorithm extends motion correction to include the whole heart. It is expected to be useful for imaging of not only the coronary arteries but also of other non-coronary intracardiac structures such as the cardiac valves.
Earlier studies that applied SSF2 reconstruction to images of the coronary arteries, of heart-and valve structures, and of the great vessels showed that the image quality was significantly improved by the algorithm and the number of non-evaluable scans was lower than of images subjected to standard-or SSF1 reconstruction 21,22 . Our study focused on the aortic annulus; it indicates that SSF2 yielded higher motion artifact correction in the whole heart. Others 30 who applied SSF1 to cardiac CT for aortic annulus measurements reported that it significantly improved the image quality of systolic CT datasets. We examined the effect of SSF2 in a wide range of heart rates and showed that it is useful for the evaluation of the aortic annulus not only in the systolic-but also in the diastolic phase. Our findings suggest that SSF2 reconstruction reduces aortic valve motion artifacts throughout the cardiac phases.
SSF2 reconstruction was not useful at R-R interval 75% in patients with a low or intermediate heart rate. At those heart rates and cardiac phases, the temporal resolution on electrocardiogram-gated scans may be sufficient and motion artifacts may not be inherent. SSF1-and SSF2 reconstruction may be useful in patients with a high heart rate and for scans with low temporal resolution 13,14,21,22 . Our findings suggest that SSF2 is as useful as SSF1 in patients with a high heart rate. SSF1 cannot address other non-coronary intracardiac structures. It calculates the movement trajectories of coronary arteries using the imaging information of the target phase and the bilateral neighboring phases to reduce motion artifacts by compensating for cardiac motion and to generate diagnosable CT images of the coronary arteries [13][14][15][16][17][18][19][20] . On the other hand, the SSF2 algorithm, a fully automated technique based on knowledge Standard-and SSF2 reconstruction at an R-R interval of 40% (A and B) and standard-and SSF2 reconstruction at an R-R interval of 75% (C and D). The solid line represents the mean difference, the dashed lines represent the 95% limit of agreement (mean difference ± 1.96 SD). SD = standard deviation. www.nature.com/scientificreports/ and feedback obtained from SSF1, seeks each region of all image volumes for a local path that is consistent with the subset of measured data. Once the vessel's motion path is identified, the data are discretized into a series of datasets based on when the corresponding projection rays were measured. Each volume dataset in the series undergoes the process of spatial deformation by the motion field. This allows the motion state to be mapped from the respective time to the central reference time that is determined by the prescribed cardiac phase 21,30 . Therefore, SSF2 can not only further reduce coronary artery artifacts due to motion in patients with high heart rates but it also improves the image quality of other cardiac vascular structures such as valves and cardiac muscles   www.nature.com/scientificreports/ on cardiovascular CT images. Our results support earlier findings 22 that, compared with SSF1, SSF2 further improved the image quality of not only coronary arteries but also of all valves and other cardiac structures. As the heart rate increased, the median ERD became significantly shorter on SSF2-than standard reconstruction images. A shorter ERD results in sharper edges 33,34,37 . Therefore, the ERD quantitatively demonstrates that SSF2 reconstruction reduced motion artifacts attributable to a high heart rate. The concept of quantifying motion artifacts via the EDR is further supported by the significant difference (p < 0.001) in our qualitative evaluation data.
We found that the dispersion of the area and perimeter of the aortic annulus measured by two observers was significantly smaller on SSF2-than standard reconstruction images regardless of the heart rate. This suggests that SSF2 reconstruction facilitates measurement of the aortic annulus by reducing motion artifacts. Although we did not assess its measurement accuracy by comparing our findings with the reference standard for transesophageal echocardiography 32 , we expect that the routine use of SSF2 will result in higher measurement accuracy than conventional standard reconstruction imaging in patients with severe aortic stenosis.
Although cardiac CT is the reference standard for the workup of TAVI candidates scheduled for an investigation of the aortic root 1,2 , motion artifacts reduce the accuracy of aortic annulus sizing and directly impact on patient outcome after TAVI procedure 2,7-9 . As a result of evaluating the dispersion between the two reconstructions with respect to the sizing of the aortic annulus, SSF2 was significantly smaller than standard regardless of the patients' heart rate or R-R interval. For TAVI planning, we still tend to use systolic imaging for the measurements 2,4,30,31 and the aortic annulus seems to be better delineated when SSF2 is used. Therefore, SSF2 may contribute to improving the accuracy of sizing of the aortic annulus. Figure 5. Results of the equivalence test. Results of the equivalence test for the difference in CNR between standard and SSF2 reconstruction. CNR = contrast-to-noise ratio; HR = heart rate; SSF2 = SnapShot Freeze 2. www.nature.com/scientificreports/ During catheter-based implantation especially of the balloon-expandable prosthesis, it is important to use a fluoroscopic projection that provides an exact orthogonal view onto the aortic annular plane 2,8,9 . Because CT offers a 3D dataset, it allows identification of appropriate projection angles that will provide an orthogonal view onto the aortic valve plane 38,39 . Others 1,2 reported that appropriate angles can be predicted from pre-TAVI cardiac CT scans when dedicated automated software programs are used. However, this is only possible if the aortic valve plane is accurately defined on the CT scan 39,40 . As SSF2 accurately delineated the aortic annulus, its application may result in a more accurate automatic processing of CT images for TAVI.
As renal dysfunction is relatively common in elderly patients scheduled for TAVI, a low-contrast protocol is recommended 41 . SSF2 reconstruction may be appropriate in TAVI candidates with renal dysfunction because it not only improves the image quality but also reduces the need for rescanning.
To avoid the potential impact of SSF2 reconstruction on quantitative measurements of the ERD, we measured the CT number in the ascending aorta, the image noise, and the CNR on SSF2 reconstructed images. We found that CNR was equivalent between scans subjected to standard-or SSF2 reconstruction irrespective of the patients' heart rate, confirming that SSF2 corrected only the motion artifacts and that it did not affect other parameters.
Our study has some limitations. First, our study population was relatively small and our investigation was based on single-center experiences. Second, because we had no reference standard such as transesophageal echocardiographs for all patients, our ability to determine whether the subjectively judged aortic annulus image www.nature.com/scientificreports/ quality improves the accuracy of measurements. Third, we only evaluated the differences between two reconstruction algorithms applied on the same CT scanner and we did not compare our findings with those made on, for example, dual-source CT scans. Finally, additional studies are underway to determine whether the robustness of SSF2 reconstruction allows lowering the preset padding range prior to scanning, thereby minimizing the required radiation dose.

Conclusions
In conclusion, our findings suggest that the SSF2 algorithm was superior to standard reconstruction because it improved the image quality and reduces motion artifacts especially in patients with a high heart rate or a 40% R-R interval. These findings may help SSF2 improve the accuracy of sizing of the aortic annulus prior to TAVI (" Supplementary Information").