Coronary computed tomography angiography (CCTA) is widely used in the diagnosis and evaluation of cardiovascular diseases, including coronary heart disease, congenital heart disease, valvular abnormalities, and atrial fibrillation [1, 2, 3, 4, 5]. Elevated heart rate increases the incidence of motion artifacts and decreases image quality. Recent studies have revealed that the heart motion correction algorithm yields satisfactory images in patients with high heart rates [6]. The second-generation snapshot freeze (SSF2) algorithm—a novel deep learning-based heart motion correction method integrated into CCTA reconstruction—has enhanced coronary artery imaging in patients with increased heart rate [7]. Furthermore, SSF2 effectively stabilizes the left atrium, left atrial appendage, heart valve, and other structures, thereby improving diagnostic capabilities [1, 5].

Heart segmentation on CCTA is widely used to analyze cardiac structure and function [8, 9]. Manual segmentation remains feasible and strongly correlates with cardiac magnetic resonance (CMR) and echocardiography in evaluating atrial and ventricular volumes and ejection fraction in patients with moderate or low heart rate [8, 9, 10]. Achieving high-quality CCTA images in patients with high heart rates remains challenging. SSF2 has improved image quality during systole more effectively than during diastole in single-phase reconstruction of the valve and left atrium in patients with high heart rates [5]. However, its applicability to whole-cardiac-cycle reconstruction for functional analysis remains uncertain.

The whole-cardiac-cycle reconstruction of CCTA enables cardiac function and myocardial strain evaluation in patients with conditions such as coronary heart disease, valvular disease, hypertrophic cardiomyopathy, atrial fibrillation, and congenital heart disease [2, 4, 11, 12, 13]. However, results from different studies have indicated considerable variability. Recent studies revealed a strong correlation between CCTA, echocardiography, and CMR in the analysis of cardiac function, myocardial strain [4, 10, 14], and prognostic value [1, 15]. However, some studies have reported discrepancies in functional and strain parameters across modalities [10, 16, 17, 18]. There is a good reproducibility of myocardial strain measurements at moderate or slow heart rates (69 $\pm$ 12 beats/min) [19]; however, the impact of high heart rates on myocardial strain remains underinvestigated. Unlike single-phase reconstruction for coronary artery assessment, whole-cardiac-cycle reconstruction encompasses the full cardiac cycle. When the heart rate is $\ge$80 beats/min, the rapid contraction and relaxation of the heart—particularly during the rapid ejection and filling phase—pose significant challenges for cardiac motion freezing. Currently, no study has assessed cardiac function using heart motion correction algorithms or SSF2 in patients with high heart rates, thereby leaving an important gap in the literature.

This study aimed to investigate the application of SSF2 in enhancing CCTA image quality and heart segmentation, as well as the assessment of cardiac structure and function and myocardial strain in patients with high heart rates.

This study enrolled 58 patients with high heart rates, with an average heart rate of 89.40 $\pm$ 7.25 beats/min (range: 80–108 beats/min). The average dose length product was 452.06 $\pm$ 41.87 mGy $\times$ cm). Supplementary Table 1 presents detailed baseline characteristics.

The image quality scores for endocardium, valves, papillary muscles, chordae tendineae, and trabeculae were higher; the overall image quality for diagnosis score was higher in SSF2 protocol. (3.91 $\pm$ 0.29 vs. 3.84 $\pm$ 0.37, p 0.01; Table 1 and Fig. 2). In terms of quantitative parameters, SNR (9.02 $\pm$ 4.54 vs. 7.30 $\pm$ 4.00, p 0.01), CNR (6.55 $\pm$ 3.62 vs. 5.03 $\pm$ 3.34, p 0.01), and ERS (41.71 $\pm$ 19.03 vs. 25.59 $\pm$ 13.16, p 0.01) were higher, whereas entropy (4.12 $\pm$ 0.48 vs. 4.40 $\pm$ 0.28, p 0.01) was lower in the SSF2 group (Fig. 3).

Fig. 2.

Image quality and artifact assessment between the STD and SSF2 groups by two readers. p = 0.03 for the comparison between groups, the remaining p-values were 0.01 between groups. R1, reader 1; R2, reader 2; 4 = excellent, no artifact; 3 = good, minimal artifact, no impact on diagnosis; 2 = some artifact in the heart, but still usable for diagnosis; 1 = poor, significant artifact, diagnosis not possible. K, kappa value.

Fig. 3.

Comparisons of ERS and entropy between the STD and SSF2 groups. All p-values were 0.01.

Automatic segmentation of the left ventricle indicated that left ventricular end-diastolic volume (LVEDV) and left ventricular end-systolic volume (LVESV) were increased in the STD and SSF2 protocols compared with the MC protocol. The ICCs for LVEDV and LVESV were excellent ($>$0.95), and the CVs of LVEDV was smaller (9%) (Table 2). The automatic segmentation of the whole heart for the STD and SSF2 protocols indicated that differences of all parameters were small, among which the mean differences in volume were 2 mL, and the mean differences of ejection fraction values were 3% between protocols (Table 3). The volume differences at 90%, 95%, and 100% phases were larger, and the average differences of left ventricular, right ventricular, left atrial, and right atrial volume variation were 2.61% (2.37%–2.75%), 3.06% (2.69%–3.51%), 4.68% (4.12%–5.62%), and 6.98% (6.23%–7.90%), respectively. However, the average differences for 0%–85% phases were 1.80% (0.97%–2.41%), 1.57% (0.82%–2.21%), 1.34% (0.84%–1.81%), and 2.31% (1.18%–3.63%), respectively. The difference in atrial volume was slightly larger than that in ventricular volume (Fig. 4).

Fig. 4.

Percentage variation of left atrial, left ventricular, right atrial, and right ventricular volumes between STD and SSF2 groups. Percentage variation of chamber volume between STD and SSF2 groups for left atrial volume (a), left ventricular volume (b), right atrial volume (c), right ventricular volume (d), respectively. Percentage variation = (Volume_SSF2 – Volume_STD) / Volume_SSF2.

The intraobserver and interobserver difference assessment revealed good myocardial strain reproducibility of Medviso Segment CT for CCTA and Circle CVI42 for CMR, and the ICCs were $>$0.85. Compared with the STD protocol, the SSF2 protocol demonstrated higher myocardial strain values (–13.26 $\pm$ 3.25 vs. –6.01 $\pm$ 1.69, p 0.01, 18.33 $\pm$ 8.11 vs. 10.60 $\pm$ 9.54, p 0.01, –11.89 $\pm$ 4.47 vs. –3.92 $\pm$ 1.89, p 0.01, and 9.89 $\pm$ 6.65 vs. 6.35 $\pm$ 3.23, p 0.01 for GCS, GRS-SA, GLS, and GRS-LA, respectively; Fig. 5 and Supplementary Table 2). The ICCs of LVEDV and LVESV among the STD, SSF2, and CMR groups were excellent. Compared with the STD protocol, the SSF2 protocol exhibited a better ICC and a smaller CV of myocardial strain with the CMR protocol. Compared with the CMR protocol, the STD and SSF2 protocols demonstrated lower GRS-SA (9.33 $\pm$ 6.85 vs. 17.93 $\pm$ 7.76 vs. 21.49 $\pm$ 9.52) and GRS-LA (6.37 $\pm$ 3.62 vs. 10.41 $\pm$ 6.66 vs. 17.01 $\pm$ 10.44) and higher GCS (–5.99 $\pm$ 1.38 vs. –13.30 $\pm$ 3.42 vs. –15.01 $\pm$ 4.44) and GLS (–4.08 $\pm$ 1.99 vs. –11.80 $\pm$ 4.83 vs. –13.01 $\pm$ 4.36). Considering differences between the STD and CMR protocols, the SSF2 and CMR protocols reached statistical significance (Table 4). The ICCs of GCS, GRS-SA, GLS, and GRS-LA were good between the SSF2 and CMR groups, and the SSF2 protocol demonstrated a smaller CV of myocardial strain with the CMR protocol.

Fig. 5.

Comparisons of strains between the STD and SSF2 groups. All p-values were 0.01.

Accurate cardiac function analysis depends on the reconstruction of the whole cardiac cycle. Incorporating additional reconstruction phases that cover the full cardiac cycle improves functional evaluation accuracy [22]. In addition, high-quality CCTA imaging is crucial for assessing subendocardial diseases, excessive trabeculation, mural thrombus, cardiomyopathy, cardiac tumors, and other conditions. Consensus documents also confirm that CCTA reliably evaluates the quantification of left and right ventricular volumes and ejection fraction, showing excellent agreement with the reference standard CMR [23]. The findings of this study demonstrate that the application of SSF2 significantly enhances image quality in patients with high heart rates and improves CNR between the myocardium and ventricle. This enhancement facilitates improved detection and differentiation of thrombus, as well as accurate assessment of volume and ejection fraction. Moreover, images obtained using SSF2 offer greater reliability for myocardial strain evaluation compared to STD protocol. These results underscore the clinical relevance of SSF2 in the assessment of cardiac structure and function among patients with high heart rate.

When the heart rate is $\ge$80 beats/min, a prolonged ejection period and shortened filling period would lead to pronounced artifacts. SSF2 has improved valve image quality in patients with high heart rates [3]. However, limited research has addressed the image quality of the endocardium and papillary muscle. High heart rate has been reported to cause endocardium blurring, inaccurate segmentation, and impaired assessment of ventricular volume and function without heart motion correction algorithm [22]. This study is the first to assess the feasibility of SSF2 for full cardiac cycle reconstruction in patients with high heart rates. Our results demonstrate that the SSF2 protocol effectively reduced motion artifacts and improved SNR and CNR of the myocardium and blood pool, resulting in a more uniform and less heterogeneous blood pool. However, in some cases, heterogeneous mixing of contrast medium and blood in the right atrium was observed, which is considered an artifact and incorrectly calibrated with SSF2. To improve the mixing of contrast medium and blood and ensure accurate assessment of cardiac structure and function, dual-bolus contrast media injection or postpositive dual-flow CCTA [24] is recommended.

Manual CCTA segmentation has demonstrated comparable efficacy to CMR in assessing left ventricular ejection fraction [10]. In our study, the difference between automatic segmentation using the SSF2 protocol and manual segmentation is smaller than that observed with the STD protocol. However, the difference in LVESV between the SSF2 and STD protocols is not statistically significant. Notably, the difference in LVESV between automatic and manual segmentation using the SSF2 protocol is statistically significant. Several factors may be associated with this. First, CCTA images predominantly comprise diastolic frames, which are used for training using the 3D U-Net algorithm. Its performance on systolic frames likely suffers due to the unbalanced training dataset. Furthermore, at end-systole, the gap between the papillary muscles, trabeculations, and myocardium is smaller or even absent. Another contributing factor may be the overtracing of the endocardial boundary and misdefinition of the mitral valve annulus plane when using the automatic approach [8].

Compared with automatic segmentation using the STD protocol, the end-diastolic volumes of left ventricle, left atrium and right atrium are larger when using the SSF2 protocol. This difference is primarily due to a clearer and sharper boundary between the endocardium and the blood pool in the SSF2 protocol. Significant differences in automatic segmentation between the SSF2 and STD protocols are observed in the 90%, 95%, and 100% phases, corresponding to the rapid ejection phases of the cardiac cycle. During these phases, the rapid contraction of the ventricles results in more artifacts, highlighting the superior motion correction performance of the SSF2 protocol in these phases.

CT strain analysis is highly variable due to various CT techniques, protocols, and post-processing algorithms [25]. Our results indicate that the strain measurements obtained using the STD and SSF2 protocols are underestimated compared with those derived from CMR. Several factors contribute to these discrepancies. First, myocardial strain analysis depends on precise endocardial tracing, and poor-quality endocardial images and suboptimal automatic segmentation during systole significantly affect results [26]. Second, the reconstructed cardiac phases and time resolution affect strain measurements. CCTA has underestimated GLS compared with speckle-tracking echocardiography [9]. Variations in the reconstruction increments of the R-R interval (5% and 10%) are a significant source of differences in LV and LA CT-FT strain values [22]. The cardiac phases in CCTA images are 21, fewer than the 30 phases in CMR. Third, differences in myocardial strain may originate from variations in imaging modalities and real-time heart rate [22, 26]. The strain results can differ depending on the post-processing software used [16]. Compared with the STD protocol, the SSF2 protocol demonstrated improved ICCs and relatively smaller CVs in comparison with CMR for myocardial strain assessment. However, the CVs for certain strain parameters, particularly GLS and GRS, remained relatively high, indicating notable variability at the individual level. These findings suggest that SSF2-derived strain measurements show good correlation and consistent trends with CMR, rather than strict interchangeability between the two modalities.

The SSF2 protocol significantly improves the image quality for whole-cardiac-cycle reconstruction and cardiac structure and function evaluation in patients with high heart rates. The results of automatic segmentation indicate strong agreement with manual segmentation for LVEDV evaluation, demonstrating considerable potential for clinical evaluation, despite poor agreement for LVESV. The SSF2 protocol outperforms the STD protocol in myocardial strain evaluation, particularly for GCS and GLS, and demonstrates improved correlation with CMR measurements. However, given the relatively high variability observed for certain strain parameters, SSF2-derived strain values should be interpreted with caution and are currently more suitable for comparative or trend-based analysis rather than direct interchangeability with CMR.

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

XC, XZ and AS designed the research study. XC and YD performed the research. JP provided administrative support and provision of study materials or patients. RW, SY and XZ collected image data and did image post processing. HX contributed to segmentation of coronary computed tomography angiography. XG and JY were responsible for statistical analysis and graphing. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

The patient informed consent was waived in this retrospective study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Ethics Committee of Affiliated Jinhua Hospital, Zhejiang University School of Medicine (approval number: 2025-317).

We would like to express our gratitude to all those who helped us during the writing of this manuscript. Thanks to all the peer reviewers for their opinions and suggestions.

This project was supported by the Medical and Health Research Project of Zhejiang Province (Grant No.2025KY1745), the key project from Jinhua Municipal Science and Technology Bureau (Grant No. 2023-1-094), and the project grant from the Jinhua Municipal Central Hospital (Grant No. JY2024-7-10).

Aiyun Sun is an employee of GE Healthcare China. However, the company had no role in the handling or conduct of the study. The author had full access to all data in the study and take full responsibility for the integrity of the data and the accuracy of the data analysis.

Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.31083/RCM48026.