Effect of SnapShot Freeze 2 on Reducing Pulsation Artifacts in Coronary Artery Imaging of Patients With High Heart Rates

Introduction

Coronary heart disease is a leading cardiovascular condition, with increasing rates of mortality and morbidity [1]. Through continuous improvements in multislice spiral computed tomography (CT) resolution, coronary computed tomography angiography (CCTA) – a noninvasive imaging modality – has become a preferred method for screening and diagnosis of coronary heart disease [2].

Pulsation artifacts in coronary artery imaging can substantially reduce the diagnostic value of CCTA. Prospective electrocardiogram (ECG)-gated CCTA is typically used to select an appropriate cardiac cycle phase based on the patient’s heart rate, thus reducing radiation exposure [3]. Although this technique can decrease the radiation dose by approximately 65% [4], it generally requires a heart rate below 70 bpm [5]. When heart rate increases during scanning, the conventional end-diastolic (70–80%) or end-systolic (40–50%) phases may not adequately suppress motion artifacts, which compromises diagnostic accuracy. Accordingly, the reduction of pulsation artifacts remains a major challenge in CCTA. Strategies such as increasing gantry rotation speed, use of dual-source CT, and administration of β-blockers have been utilized [6], but motion artifacts continue to compromise image interpretation.

In recent years, GE Healthcare introduced the SnapShot Freeze (SSF) algorithm to reduce motion artifacts in CCTA. Previous studies have demonstrated that SSF improves image quality [7,8]. However, pulsation artifacts still affect image clarity and diagnostic reliability, particularly among patients with coronary calcification or elevated heart rates [9].

SSF reduces motion artifacts by acquiring information about coronary artery motion between adjacent phases within a single cardiac cycle and compensating for residual motion at the target phase [10–13]. Nevertheless, several studies have shown persistent misdiagnoses and missed diagnoses among patients with high heart rates due to unavoidable pulsation artifacts when SSF is used [12–14]. To address these limitations, SnapShot Freeze 2 (SSF2) was developed as an automated whole-heart motion-correction algorithm that performs comprehensive motion analysis and correction [15]. Unlike SSF, SSF2 extends motion correction to the entire heart, which may improve imaging performance in patients with high or irregular heart rates. However, its effectiveness relative to SSF and conventional reconstruction has not been systematically evaluated.

In this study, CCTA images from patients with heart rates above 80 bpm were reconstructed using 3 methods – no-SSF, SSF, and SSF2 – to compare their abilities to reduce pulsation artifacts. Coronary arteries were divided into 18 segments according to the Society of Cardiovascular Computed Tomography 18-segment model. Subjective assessment was performed for each segment to evaluate image quality. Additionally, an artificial intelligence (AI)–assisted diagnostic system based on deep learning was used to determine whether SSF2 improves diagnostic accuracy compared with SSF and no-SSF reconstruction.

Material and Methods

ETHICS APPROVAL:

The study protocol was reviewed and approved by the Institutional Review Board of West China Hospital (approval number: 2022 Nian Shen [611]). The requirement for written informed consent was waived because the study involved retrospective analysis of anonymized data.

STUDY POPULATION:

This retrospective single-center study was conducted in the Department of Radiology, West China Hospital, Sichuan University. Consecutive patients who underwent clinically indicated CCTA between January and June 2023 were screened for inclusion.

INCLUSION AND EXCLUSION CRITERIA:

Adults (age ≥18 years) with a heart rate above 80 bpm at image acquisition who completed CCTA during the study period were eligible. Patients were excluded if they had (1) arrhythmia or irregular heart rhythm that interfered with ECG gating; (2) a history of coronary revascularization procedures (stent implantation or coronary artery bypass grafting); (3) severe coronary artery calcification that obscured lumen visualization; (4) a known allergy or contraindication to iodinated contrast media; (5) renal insufficiency (estimated glomerular filtration rate < 60 mL/min/1.73 m2); (6) inability to comply with breath-hold instructions or poor image quality due to motion; or (7) non-diagnostic or incomplete CCTA data after reconstruction. These criteria were used to minimize confounding factors that could influence image quality or motion-artifact severity and to ensure comparability among reconstruction algorithms. The patient selection process, exclusions, and final analytic groups are summarized in Figure 1.

STUDY SAMPLE AND GROUPING:

Patients were categorized by acquisition heart rate into two descriptive strata (80 to 90 bpm and greater than 90 bpm). No statistical subgroup comparisons were performed; all analyses were conducted in the overall high–heart-rate cohort.

DATA HANDLING:

For each patient, both end-diastolic (70–80%) and end-systolic (40–50%) phases were reconstructed when available. Images not meeting diagnostic requirements after standard reconstruction were excluded from phase-specific comparisons but were recorded in the study flowchart. No imputation was performed for missing or excluded datasets.

SAMPLE-SIZE RATIONALE:

The final sample size (n=50) was determined by the number of eligible high–heart-rate patients during the study period; it was consistent with prior investigations of SSF and SSF2 that identified statistically significant differences in artifact-reduction indices using comparable cohorts. Although no formal power analysis was conducted, this sample size was considered sufficient for exploratory quantitative comparisons among reconstruction methods.

CT SCANNING PARAMETERS:

All CCTA examinations were performed with a GE Revolution CT scanner (GE Healthcare, Waukesha, WI, USA). The scanning range extended from 1 cm below the tracheal carina to 2 cm below the diaphragm in the craniocaudal direction. Prospective ECG gating was used with the following parameters: collimation, 128×0.625 mm; detector coverage, 160×0.625 mm; reconstruction slice thickness, 0.625 mm; slice interval, 0.625 mm; gantry rotation time, 0.28 s per rotation; maximum tube current, 400–900 mA with automatic modulation; and tube voltage, 120 kV.

CONTRAST MEDIUM ADMINISTRATION:

Iopamidol (400 mg I/mL; Bracco, Milano, Italy) was administered through an antecubital vein with a high-pressure injector. Both contrast medium and normal saline were delivered using a dual-head injector. Scanning was triggered automatically at the descending aorta near the tracheal bifurcation under bolus-tracking guidance. Monitoring began 9 s after injection; scanning was initiated when the attenuation threshold reached 110 Hounsfield units, with a 3.5-s delay and no breath-hold voice prompt. The injection rate was 5 mL/s for a total contrast volume of 75 mL, followed by 30 mL of normal saline at the same rate.

RADIATION DOSE INDICES:

The computed tomography dose index (CTDIvol, mGy) and dose-length product (DLP, mGy·cm) were automatically recorded by the scanner to quantify radiation exposure. CTDIvol represents the mean dose delivered per slice, whereas DLP reflects total scan exposure adjusted for scan length. Both metrics were used as indicators of radiation efficiency and patient safety.

DATA INTEGRITY AND HANDLING:

All images were reconstructed for both end-diastolic (70–80%) and end-systolic (40–50%) phases. Images not meeting diagnostic requirements due to severe motion or incomplete reconstruction were excluded from quantitative comparisons but were documented in the study flowchart (Figure 1). No data imputation was performed. When a phase was non-diagnostic, only the evaluable phase was retained for statistical analysis to ensure data reliability.

AI SOFTWARE AND EVALUATION SETTINGS:

Automated coronary artery analysis was performed using the United Imaging CCTA Analysis System (uAI–CoronaryCTA 1.0; United Imaging Healthcare, Shanghai, China). This software utilizes a deep learning–based coronary segmentation and labeling algorithm trained on a multi-vendor dataset of more than 1000 annotated CCTA examinations; its performance has been externally validated elsewhere [16].

CROSS-VENDOR WORKFLOW AND COMPATIBILITY:

Although CCTA acquisition was performed on a GE Revolution CT system, all image data were exported in DICOM format compatible with the United Imaging platform. This cross-vendor workflow was required because SSF and SSF2 are vendor-specific reconstruction algorithms, whereas the United Imaging software functions as a vendor-neutral post-processing tool. Consequently, the present findings primarily reflect the performance of GE-based reconstruction methods and may not be fully generalizable to other CT vendors.

AI-DRIVEN DATASETS:

Each of the 50 patients contributed 2 reconstructed cardiac phases (end-diastolic and end-systolic), resulting in 100 phase-level datasets. Each dataset underwent 3 motion-correction reconstructions (no-SSF, SSF, and SSF2). The AI system automatically identified coronary artery branches and quantified the number of detectable segments for each reconstruction method.

INTERPRETATION FRAMEWORK:

To provide a reference standard, all AI outputs were qualitatively compared with blinded, independent readings by 2 radiologists. Because the study primarily focused on optimizing image quality and reducing motion artifacts, rather than diagnostic accuracy, AI evaluation was included as a secondary exploratory endpoint that complemented subjective and objective image-quality assessments. Accordingly, sensitivity, specificity, and area under the curve metrics were not calculated. The AI analysis was used to quantify improvements in coronary segment detectability, rather than to assess clinical diagnostic performance.

IMAGE POST-PROCESSING AND ANALYSIS:

Images were reconstructed at the 70% to 80% and 40% to 50% phases. Images not meeting diagnostic requirements were retrospectively reconstructed by experienced radiologists using the acquired data. For optimal phases, images were reconstructed with the SSF and SSF2 algorithms, respectively. Reconstructed images were classified into 3 groups: no-SSF, SSF, and SSF2. Image quality was compared across these groups via subjective assessment, objective measurement, and AI-enabled evaluation. After application of the 2 motion-freezing methods to the original data from 50 patients, 279 image sequences were generated using SSF and SSF2 reconstructions. The research workflow is shown in Figure 2.

SUBJECTIVE EVALUATION:

Two radiologists independently assessed image quality for 18 segments of the 3 major coronary arteries (right coronary, left circumflex, and left anterior descending) using a blinded design and the Society of Cardiovascular Computed Tomography 18-segment model [17]. The 18 coronary artery segments were defined as follows: (1) proximal right coronary artery; (2) mid right coronary artery; (3) distal right coronary artery; (4) right posterior descending artery; (5) left main; (6) proximal left anterior descending artery; (7) mid left anterior descending artery; (8) distal left anterior descending artery; (9) diagonal branch 1; (10) diagonal branch 2; (11) proximal circumflex artery; (12) obtuse marginal branch 1; (13) mid and distal left circumflex artery (LCx); (14) obtuse marginal branch 2; (15) posterior descending artery from the LCx; (16) right posterolateral branch; (17) ramus intermedius; and (18) posterolateral branch from the LCx. Image quality was graded using a 5-point Likert scale [18]. A score of 5 indicated excellent image quality, characterized by a continuous and intact coronary lumen, sharp margins, and absence of motion artifacts. A score of 4 indicated a continuous and intact lumen with mild localized blurring and no evident motion artifacts. A score of 3 indicated reduced sharpness or mild motion artifacts without diagnostic impact. A score of 2 indicated moderate motion artifacts or partial vessel obscuration. A score of 1 indicated poor image quality, characterized by coronary misalignment, severe artifacts, and inability to support diagnosis. Images with scores of 3 or higher were considered diagnostically acceptable.

All images were evaluated independently by 2 radiologists who were blinded to reconstruction technique and patient information. Interobserver agreement was assessed using Cohen’s kappa statistics, yielding a kappa value of 0.78.

OBJECTIVE EVALUATION:

Image slices with the most severe artifacts – particularly in the right coronary, left anterior descending, and left circumflex arteries – were selected as regions of interest. Areas with no or minimal artifacts served as control regions. Standard deviation (SD) values of these regions were measured. The artifact index was calculated using the following formula:

AI-ENABLED EVALUATION:

The United Imaging CCTA analysis system automatically identified coronary artery branches in the no-SSF, SSF, and SSF2 groups. The number of coronary artery segments detected for images processed with each algorithm was recorded. AI analysis was performed using the United Imaging CCTA Analysis System (Plus v1.0; United Imaging Healthcare). The performance of deep learning–based coronary artery segmentation and labeling in CCTA has been externally validated in independent studies, revealing robust accuracy across heterogeneous datasets [19]. Data acquired via the GE Revolution CT system were exported in DICOM format compatible with the United Imaging platform.

STATISTICAL ANALYSIS:

Statistical analyses were performed using SPSS version 26.0. The Shapiro-Wilk test was utilized to assess data normality. Normally distributed variables are expressed as mean±standard deviation, whereas non-normally distributed variables are reported as median and interquartile range. Analysis of variance was used to compare the 3 groups, followed by Bonferroni correction for pairwise comparisons. The threshold for statistical significance was regarded as P<0.05.

Although no a priori power analysis was performed, post hoc sensitivity analysis was conducted to estimate the minimum detectable effect with the current sample size. For repeated-measures comparisons among the 3 reconstruction algorithms (no-SSF, SSF, and SSF2) with n=50, α=0.05, and power (1-β)=0.80, the design provided approximately 80% power to detect a medium effect size (Cohen’s f≈0.25), corresponding to a pairwise Cohen’s dz of approximately 0.40. This effect size translates to a detectable difference of approximately 0.3–0.4 points on the 5-point Likert image-quality scale (SD≈0.8–1.0) or 0.8–1.0 segments for AI-based detection (SD≈2.0–2.5). These estimates indicate that the current sample size was sufficient to detect clinically relevant differences in image-quality metrics across reconstruction algorithms.

Results

PATIENT DEMOGRAPHICS AND IMAGING CHARACTERISTICS:

The analysis included 50 patients (26 men and 24 women; mean age, 59.35±13.01 years; range, 34–78 years). The mean heart rate during CCTA acquisition was 86.4±5.9 bpm (95% confidence interval [CI], 84.8–88.0). All examinations were successfully completed without adverse contrast reactions or missing datasets. Radiation dose parameters were CTDIvol, 38.6±5.1 mGy (95% CI, 37.0–40.2) and DLP, 524.8±62.3 mGy·cm (95% CI, 507.1–542.5), indicating acceptable exposure within institutional safety limits. No significant differences in radiation dose were observed among reconstruction algorithms (P=0.72).

SUBJECTIVE IMAGE-QUALITY EVALUATION:

Significant differences in subjective image-quality scores were observed among the 3 reconstruction algorithms in both diastolic and systolic phases (F=21.0–32.4; all P<0.001). Post hoc Bonferroni analyses showed that SSF2 achieved significantly higher subjective scores than SSF and no-SSF across all coronary territories (all pairwise P<0.001). The consistent ranking of SSF2 >SSF >no-SSF was maintained in both cardiac phases, indicating superior motion-compensation performance with SSF2. Detailed results are presented in Tables 1 and 2.

OBJECTIVE ARTIFACT-INDEX ANALYSIS:

Significant differences in artifact index values were observed among the 3 reconstruction algorithms in both cardiac phases (F=6.96–25.65; all P≤0.001). Compared with SSF and no-SSF, SSF2 consistently produced the lowest artifact indices, reflecting reduced motion-artifact severity. The extent of artifact-index reduction was greater in the diastolic phase than in the systolic phase. Detailed comparisons are summarized in Table 3.

AI-ENABLED EXPLORATORY EVALUATION:

AI-based quantitative analysis demonstrated significant differences among reconstruction algorithms in the number of automatically identified coronary artery segments (F=14.23; P<0.001). Post hoc comparisons confirmed that SSF2 detected significantly more analyzable coronary segments than SSF and no-SSF (all P<0.001). These findings were consistent with trends observed in subjective and objective evaluations; they indicate improved interpretability of images processed with SSF2.

Discussion

LIMITATIONS:

This study has several limitations. First, it was a single-center retrospective analysis with a relatively small sample size (n=50), which may constrain statistical power and generalizability. Second, potential selection bias cannot be excluded because only patients with high heart rates were included. Third, although 2 independent radiologists performed blinded image evaluations, residual observer bias remains possible. Fourth, the analysis focused on short-term image optimization without assessing long-term diagnostic accuracy or clinical outcomes. Fifth, the exploratory AI-based analysis was not validated for diagnostic sensitivity, specificity, or area under the curve. Finally, because SSF and SSF2 are vendor-specific reconstruction algorithms and the AI software functions as a vendor-neutral post-processing tool, cross-vendor compatibility and reproducibility require confirmation in future multicenter studies.

Conclusions

Compared with no-SSF and SSF, images processed using the SSF2 algorithm demonstrated superior performance concerning motion artifact reduction in coronary artery imaging of patients with high heart rates during CCTA. This approach effectively reduces missed and false diagnoses related to image-quality limitations in both manual and AI-assisted interpretation, providing substantial clinical value. Although the present study evaluated the performance of AI-assisted diagnostic software in identifying coronary artery segments across different reconstruction algorithms, diagnostic accuracy under these conditions requires further investigation. Additionally, this study focused on the effects of reconstruction algorithms on motion-artifact reduction in coronary artery imaging. The impacts of no-SSF, SSF, and SSF2 on motion artifacts in cardiac valve and myocardial imaging warrant further study.