Abstract
Background
Cardiac magnetic resonance (CMR) imaging is an important tool for assessing the cardiac structure and function in Kawasaki disease. To perform a prospective analysis comparing the newly developed temporal artificial intelligence (AI)-assisted compressed sensing cine (ACS-cine) and conventional cine (Conv-cine) in the assessment of cardiac function in Kawasaki disease patients.
Methods
This study included 45 patients (35 males), who were scanned using both ACS-cine and Conv-cine sequences. Left and right ventricular ejection fractions (LVEF and RVEF), left and right ventricular end-systolic volumes (LVESV and RVESV), and left and right ventricular end-diastolic volumes (LVEDV and RVEDV) were measured. The image quality and intraclass correlation coefficient (ICC) of the two methods were also assessed.
Results
The mean acquisition time of ACS-cine was significantly shorter than that of Conv-cine (17.18±2.45 vs. 251.12±38.01 seconds, P<0.05). In terms of cardiac function measurements, the median LVEF measured by Conv-cine was higher than that of ACS-cine (63.0% vs. 61.0%, P<0.05), while median LVESV was lower in Conv-cine than in ACS-cine (26.5 vs. 27.9 mL/m2, P<0.05). The image quality score of ACS-cine was slightly higher than that of Conv-cine (4.41±0.50 vs. 3.87±0.65, P<0.001). The inter-observer consistency was almost perfect for both ACS-cine (ICC: 0.882–0.978) and Conv-cine (ICC: 0.952–0.991).
Conclusions
ACS-cine technology demonstrates significant time efficiency advantages in CMR with satisfactory image quality, although there are statistically significant differences in the measurement of certain cardiac function parameters (LVEF and LVESV) compared to Conv-cine.
Keywords: Cardiac magnetic resonance imaging (CMR imaging), compressed sensing (CS), cine sequence, Kawasaki disease, children
Highlight box.
Key findings
• Artificial intelligence-assisted compressed sensing cine (ACS-cine) technology demonstrates significant time efficiency advantages in cardiac magnetic resonance (CMR) with satisfactory image quality.
What is known and what is new?
• The multiple breath-holding method of conventional cine (Conv-cine) sequence is both time-consuming and challenging for children, for cine sequence in children, compressed sensing technology has been proven to be able to complete scanning with only a few breath-holds or during free breathing, significantly improving the feasibility and comfort for the children.
• In our research center, we used ACS-cine to achieve fast whole-heart cine scanning in free breathing for children, by means of motion correction to reduce respiratory artifacts.
What is the implication, and what should change now?
• This study aims to compare the performance of ACS-cine and Conv-cine techniques in the assessment of cardiac function in children with Kawasaki disease, focusing on exploring the potential of ACS-cine in reducing acquisition time, minimizing breath-holding requirements, and maintaining image quality.
• Future studies can expand the sample size and include more types of pediatric cardiac disease to verify the general applicability of ACS-cine technology.
Introduction
Cardiac magnetic resonance (CMR) imaging is an important technique for assessing cardiac structure and function, playing a crucial role in the diagnosis and treatment of complex cardiac diseases. Traditional CMR techniques, such as balanced steady-state free precession (bSSFP) cine sequences, can provide high-resolution dynamic cardiac images. However, their long acquisition time and high demands for patient breath-holding limit their clinical application, especially in pediatric patients or those with limited cardiopulmonary function (1). Since the multiple breath-holding method of conventional cine (Conv-cine) sequence is both time-consuming and challenging for children, achieving faster cine imaging has been a research hotspot. The compressed sensing (CS) technology has provided a new solution for accelerating CMR. By using sparse sampling and iterative reconstruction algorithms, it can maintain high image quality while reducing the amount of data acquisition, thereby significantly shortening the scanning time (2-4). For cine sequence in children, CS technology has been proven to be able to complete scanning with only a few breath-holds or during free breathing, significantly improving the feasibility and comfort for the children (5,6).
The application of artificial intelligence (AI), machine learning, and deep learning methods to assist CS-cine imaging further enhances imaging speed, image quality, and clinical practicality (7,8). A study by Yan et al. showed that high-quality cardiac cine images can be reconstructed at extremely low sampling rates using deep learning algorithms. However, that study required patients to cooperate with breath-holding, greatly increasing the difficulty of scanning (9). For pediatric patients, there is an urgent need for cardiac cine scanning under free breathing. Therefore, the development of efficient and accurate imaging techniques is of great significance. In our research center, we used AI-assisted CS cine (ACS-cine) to achieve fast whole-heart cine scanning in free breathing for children, by means of motion correction (MOCO) to reduce respiratory artifacts (10,11). ACS-cine achieves whole-heart cine imaging in children within 20 seconds of free breathing through an ultra-high acceleration factor, significantly reducing the time required for pediatric cardiac scans and increasing the success rate (10,11).
Kawasaki disease is a systemic vasculitis that mainly affects children and often leads to coronary artery lesions (CALs) and cardiac dysfunction. Accurate assessment of cardiac function in patients with Kawasaki disease is crucial for guiding treatment and prognostic evaluation (12). The application of Conv-cine in Kawasaki disease is limited by its long acquisition time, while ACS-cine technology may provide a more efficient imaging solution. This study aims to compare the performance of ACS-cine and Conv-cine techniques in the assessment of cardiac function in children with Kawasaki disease, focusing on exploring the potential of ACS-cine in reducing acquisition time, minimizing breath-holding requirements, and maintaining image quality. We present this article in accordance with the STARD reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0641/rc).
Methods
Study subjects
This was a prospective research of consecutive patients who underwent clinical CMR in the authors’ institution (Children’s Hospital of Fudan University) between January 2024 and August 2024. Inclusion criteria were: (I) children with a confirmed diagnosis of Kawasaki disease based on American Heart Association (AHA) criteria; (II) chronic phase of disease (≥6 months from acute onset); (III) presence of CALs defined and classified according to the 2017 AHA guidelines based on Z-scores obtained from echocardiography at the time of CMR examination (dilatation: Z-score 2 to <2.5; Small aneurysm: Z-score 2.5 to <5; medium aneurysm: Z-score 5 to <10 and absolute dimension <8 mm; Giant aneurysm: Z-score ≥10 or absolute dimension ≥8 mm) (13); and (IV) clinical indication for CMR including ischemic risk assessment or routine follow-up of known CALs. Exclusion criteria were: (I) patients in the acute phase of Kawasaki disease; (II) those with contraindications to CMR (metal implants); (III) those unable to cooperate with the examination due to claustrophobia; (IV) those with severe motion artifacts affecting image quality; and (V) patients with significant arrhythmias that may affect CMR acquisition or image quality. A total of 56 patients with Kawasaki disease were initially considered for inclusion. Of these, 11 patients were excluded: 6 patients had contraindications to CMR (metal implants), 3 patients were unable to complete the examination due to claustrophobia, and 2 patients excluded for severe motion artifacts affecting both ACS-cine and Conv-cine sequences were 3 and 4 years old, attributed to patient non-cooperation during scanning despite sedation; Consequently, no diagnostic images were obtainable from either sequence for image quality assessment or cardiac function measurement. The final analysis included 45 patients. Each patient was scanned once and included only once in the analysis. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Children’s Hospital of Fudan University (No. IRB-2023-X12). Written informed consent was obtained from parents or legal guardians for all patients.
CMR imaging equipment and sequences
All CMR examinations were performed on a 3.0T MR scanner (uMR 790, United Imaging Healthcare, Shanghai, China) using a 24-channel super-flex body coil and electrocardiogram triggering. Each patient was scanned with both Conv-cine and ACS-cine sequences, in free breathing. The oral chloral hydrate/dexmedetomidine sedation protocol was used for the 11 pediatric cases, confirming that all maintained spontaneous breathing without respiratory depression or significant artifact introduction. The scan parameters were shown in Table 1. The scanning positions of the two sequences were kept consistent to ensure the comparability of the data.
Table 1. Imaging parameters.
| Sequence type | Conv-cine | ACS-cine |
|---|---|---|
| ECG pattern | Retrospective ECG-gating | Retrospective ECG-gating |
| FOV Imaging area (mm2) | 360×260 | 360×260 |
| Spatial resolution (mm3) | 1.07×1.07×6.00 | 1.25×1.25×6.00 |
| Image matrix | 224×224 | 192×192 |
| Number of slices | 12 | 12 |
| TR/TE (milliseconds) | 3.28/1.58 | 3.07/1.43 |
| Flip angle (degrees) | 50 | 55 |
| Bandwidth (Hz/pixel) | 1,000 | 1,000 |
| Number of cardiac phases reconstructed | 25 | 25 |
| Number of breath-holds | 0 | 0 |
| Scan time (seconds) | 251.12±38.01 | 17.18±2.45 |
| Total acceleration factor | 1.73 | 11.1 |
Scan time that follows a normal distribution is presented as mean ± SD. ACS-cine, artificial intelligence-assisted compressed sensing cine; Conv-cine, conventional cine; ECG, electrocardiographic; FOV, field of view; SD, standard deviation; TE, echo time; TR, repetition time.
Image acquisition and reconstruction
Conv-cine sequence: a two-dimensional (2D)-bSSFP sequence was used, and the image reconstruction employed the traditional Fourier transform method. All patients performed free breathing with 4 times signal averages during imaging.
ACS-cine sequence (14,15): the ACS-cine methodology, which is an online reconstruction cine provided by the magnetic resonance imaging (MRI) manufacturer (United Imaging Healthcare, Shanghai, China), consists of three key components: (I) pulse sequence: ACS-cine is based on a bSSFP sequence, which provides high signal-to-noise ratio and T2/T1-weighted contrast ideal for cardiac imaging. (II) Spatiotemporal undersampling: the sequence employs a variable-density Latin hypercube sampling pattern across k-space with an acceleration factor of 11.1, significantly reducing data acquisition time compared to conventional fully sampled approaches. (III) Deep learning reconstruction and MOCO: a convolutional recurrent neural network reconstructs images from undersampled data by leveraging temporal information from adjacent cardiac phases. Online MOCO is applied to reduce respiratory motion artifacts associated with free breathing. The detailed engineering rationale (Latin hypercube implementation, echo-sharing, fluctuation minimization, self-avoidance) is provided in Appendix 1.
Standard breath-held Conv-cine was not systematically performed in this cohort because many younger children and those with clinical symptoms were unable to achieve repeatable breath-holds. Both sequences were acquired during free breathing to ensure a fair comparison of the acceleration and MOCO capabilities of ACS-cine. The decision to use free-breathing imaging was based on age, sedation status and patient cooperation. Using breath-hold Conv-cine as the comparator would introduce a confounding factor (breath-hold vs. free breathing) that would make it impossible to isolate the effects of the CS and deep learning reconstruction from the effects of respiratory motion. The free-breathing Conv-cine with signal averaging serves as an appropriate reference standard for evaluating ACS-cine performance under similar physiological conditions.
Cardiac function parameter measurements
Two experienced cardiac radiologists (observer 1: X.H., 25 years of experience in CMR; observer 2: X.Z., 5 years of experience in CMR) independently measured all cardiac function parameters and assessed image quality. The final reported values represent the mean of the two independent measurements. Inter-observer agreement was assessed using intraclass correlation coefficient (ICC) for all parameters.
Cardiac function parameters were measured using commercially available post-processing software (uOmnispace.MR; United Imaging Healthcare). Papillary muscles/trabeculations were included in the ventricular cavity, right ventricular outflow tract (RVOT) was included below the pulmonary valve, and basal slices were selected based on the presence of >50% surrounding myocardium. End-diastolic and end-systolic phases were automatically detected by the software based on maximum and minimum ventricular volumes, respectively (Figure 1).
Figure 1.
Methodology for contouring the endocardial and epicardial layers of the left ventricle at the basal (A,B), mid-ventricular (C,D), and apical (E,F) levels (end-diastolic). The endocardial border (red line) was traced along the inner edge of the myocardium, while the epicardial border (green line) was traced along the outer edge of the myocardium. The papillary muscles were included in the ventricular cavity for left ventricle volume measurements. ACS-cine, artificial intelligence-assisted compressed sensing cine; Conv-cine, conventional cine.
Automated contours manual adjustment was required in 13.3% (6/45) of ACS-cine cases and 22.2% (10/45) of Conv-cine cases. The following cardiac function parameters were measured: left ventricular ejection fraction (LVEF), left ventricular end-diastolic volume (LVEDV), left ventricular end-systolic volume (LVESV), right ventricular ejection fraction (RVEF), right ventricular end-diastolic volume (RVEDV), and right ventricular end-systolic volume (RVESV). Cardiac function results were all normalized by body surface area.
Image quality assessment
Image quality was assessed by the same two cardiac radiologists who measured the cardiac function parameters, who were unaware of the CMR acquisition protocols. All image headers and sequence identification tags were removed. Both observers independently evaluated all images in a randomized order. Randomization was performed, and the two review sessions were separated by a 2-week washout interval to avoid recall bias. In cases of disagreement, the consensus reading approach was used for analysis to ensure conservative quality assessment. The research adopted a 5-point Likert scale (9). The scoring criteria are as follows:
1: non-diagnostic, the images are clinically useless;
2: poor, the images have obvious motion-related artifacts or image distortion, but some are of diagnostic value;
3: fair, the images have moderate motion-related artifacts or image distortion, but are sufficient for ventricular contour tracking;
4: good, the images have slight motion-related artifacts or image distortion;
5: excellent, the images have almost no motion-related artifacts or image distortion.
Statistical analysis
All data were statistically analyzed by SPSS 25.0, figures were generated using GraphPad Prism 9. Because this study was a small sample size study, the Shapiro-Wilk test was used to test the normality of this study data. Continuous variables are expressed as mean ± standard deviation (SD) or median [interquartile range (IQR)], and categorical variables are expressed as frequency and percentage. Among these parameters, heart rate, data acquisition time, right ventricular ventricle end-diastolic volume and RVESV all followed a normal distribution. For normally distributed paired variables between ACS-cine and Conv-cine measurements, paired samples t-tests were used (RVEDV and RVESV). For non-normally distributed paired variables, Wilcoxon signed-rank tests were applied (LVEDV, LVESV, LVEF, and RVEF). Bland-Altman analysis was used to evaluate and compare the consistency of biventricular volume and function measured by different sequences. Inter-observer consistency was assessed using ICC. The strength of agreement is as follows: 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, almost perfect agreement (16). Inter-observer agreement for image quality scoring was assessed using weighted kappa statistics. A P value <0.05 was considered statistically significant.
Results
Participant characteristics
This study included 45 patients with Kawasaki disease, and the basic information is shown in Table 2. The median age of the 45 patients was 7.6 years (IQR, 5–10 years). The median height, weight, and body surface area were 127.2 cm (IQR, 108–142 cm), 31.4 kg (IQR, 18–35 kg), and 1.07 m2 (IQR, 0.78–1.18 m2), respectively. Their mean heart rate was 83.07±10.98 beats per minute. The mean acquisition time for ACS-cine was 17.18±2.45 seconds (total cine acquisition), while the mean acquisition time for Conv-cine was 251.12±38.01 seconds (total cine acquisition). The flowchart of participant inclusion and exclusion is shown in Figure 2.
Table 2. Patient’s clinical characteristics.
| Clinical information | Data (n=45) |
|---|---|
| Gender | |
| Male | 35 (77.78) |
| Female | 10 (22.22) |
| Age (years) | 7.6 [5–10] |
| Height (cm) | 127.2 [108–142] |
| Weight (kg) | 31.4 [18–35] |
| Body surface area (m2) | 1.07 [0.78–1.18] |
| Heart rate (bpm) | 83.07±10.98 |
| NYHA class | |
| I | 40 (88.89) |
| II | 2 (4.44) |
| III | 2 (4.44) |
| IV | 1 (2.22) |
| NT-proBNP (pg/mL) | 96.33 [26.90–110.50] |
| Echocardiography LVEF (%) | 68.64 [66.00–72.00] |
| Regional wall motion abnormalities | 4 (8.89) |
| Sedation used during CMR | 11 (24.44) |
| Coronary artery dilatation | 2 (4.44) |
| Small coronary artery aneurysm | 12 (26.67) |
| Medium coronary artery aneurysm | 15 (33.33) |
| Giant coronary artery aneurysm | 29 (64.44) |
| Previous bypass surgery | 7 (15.56) |
| Heart failure | 5 (11.11) |
Values are presented as median [IQR] for non-normally distributed variables and n (%) for statistics. CMR, cardiac magnetic resonance; IQR, interquartile range; LVEF, left ventricular ejection fraction; NT-proBNP, N-terminal pro-brain natriuretic peptide; NYHA, New York Heart Association.
Figure 2.
The flowchart of participant inclusion and exclusion. ACS-cine, artificial intelligence-assisted compressed sensing cine; Conv-cine, conventional cine; KD, Kawasaki disease.
Among the 45 patients, 5 children (11.11%) had heart failure [New York Heart Association (NYHA) functional class II–IV equivalent], with heart rates ranging from 92 to 120 beats per minute (mean: 104.2±10.4 bpm). All 5 patients successfully completed both the ACS-cine and Conv-cine examinations without complications. Both ACS-cine and Conv-cine consistently detected regional wall motion abnormalities in all 4 patients who presented with regional wall motion abnormalities (100% concordance). All 45 patients had CALs. The distribution of CALs was: coronary artery dilatation in 2 patients (4.44%), Small Coronary artery aneurysm in 12 patients (26.67%), Medium Coronary artery aneurysm in 15 patients (33.33%), and Giant Coronary artery aneurysm in 29 patients (64.44%). Seven patients (15.56%) had previously undergone coronary artery bypass grafting.
Cardiac function parameter
The results of cardiac function parameter measurements are shown in Table 3. In our cohort of 45 patients, 5 patients (11.1%) had reduced LVEF (<55%; range, 38–53%). There were significant differences between ACS-cine and Conv-cine in LVEF and LVESV (P<0.05). Specifically, the median LVEF measured by Conv-cine was slightly higher than that of ACS-cine (63.0% vs. 61.0%, P<0.05), while the median LVESV was lower (26.5 vs. 27.9 mL/m2, P<0.05). The median LVEDV measured by ACS-cine was slightly lower than that of Conv-cine (73.1 vs. 74.0 mL/m2), but the difference was not statistically significant. In the measurement of RVEF, RVEDV, and RVESV, there was no significant difference between the two techniques, but the values measured by ACS-cine were slightly higher than those of Conv-cine.
Table 3. Cardiac function measurements using ACS-cine and Conv-cine.
| Items | ACS-cine | Conv-cine | P value |
|---|---|---|---|
| LVEF (%) | 61.0 (56.2–65.1) | 63.0 (57.6–66.5) | 0.02 |
| LVEDV (mL/m2) | 73.1 (67.5–79.5) | 74.0 (64.4–81.2) | 0.48 |
| LVESV (mL/m2) | 27.9 (24.3–33.5) | 26.5 (23.0–30.7) | 0.01 |
| RVEF (%) | 51.3 (48.0–57.4) | 50.9 (48.3–58.8) | 0.96 |
| RVEDV (mL/m2) | 73.4±14.3 | 72.3±13.8 | 0.19 |
| RVESV (mL/m2) | 34.4±9.1 | 34.0±8.6 | 0.42 |
Values are presented as mean ± SD for normally distributed variables and as median (IQR) for non-normally distributed variables. ACS-cine, artificial intelligence-assisted compressed sensing cine; Conv-cine, conventional cine; IQR, interquartile range; LVEDV, left ventricular end-diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end-systolic volume; RVEDV, right ventricular end-diastolic volume; RVEF, right ventricular ejection fraction; RVESV, right ventricular end-systolic volume; SD, standard deviation.
Bland-Altman analysis revealed the following bias and 95% limits of agreement (LoA): LVEF: −0.97 (95% LoA: −5.62 to 3.68); LVEDV: 0.51 (95% LoA: −8.98 to 10.00); LVESV: 0.93 (95% LoA: −3.97 to 5.65); RVEF: 0.01 (95% LoA: −3.85 to 3.87); RVEDV: 1.06 (95% LoA: −11.68 to 13.80); RVESV: 0.48 (95% LoA: −5.91 to 6.87). The Bland-Altman scatter plot showed that there was a low bias between the results of ACS-cine and Conv-cine, and overall, the agreement was acceptable without any obvious systematic trend (Figure 3).
Figure 3.
Bland-Altman plots showing agreement between ACS-cine and Conv-cine for LVEF (A), LVEDV (B), LVESV (C), RVEF (D), RVEDV (E), and RVESV (F). The Y-axis represents ACS-cine minus Conv-cine values. The solid line indicates mean bias, and the dashed lines indicate the 95% LoA. ACS-cine, artificial intelligence-assisted compressed sensing cine; Conv-cine, conventional cine; LoA, limits of agreement; LVEDV, left ventricular end-diastolic volume; LVEF, left ventricular ejection fraction; LVESV, left ventricular end-systolic volume; RVEDV, right ventricular end-diastolic volume; RVEF, right ventricular ejection fraction; RVESV, right ventricular end-systolic volume; SD, standard deviation.
Inter-observer consistency was almost perfect for both ACS-cine (ICC range, 0.882–0.978) and Conv-cine (ICC range, 0.952–0.991) across all cardiac function parameters.
Image quality
All images showed good myocardial-blood pool contrast. The percentage of different scores were as follows: score 1 (ACS-cine, Conv-cine at 0%, 0%); score 2 (0%, 0%); score 3 (2.2%, 33.3%); score 4 (60%, 51.1%); and score 5 (37.8%, 15.6%). The detailed score distribution was shown in Figure 4. The statistical comparison between ACS-cine and Conv-cine remained highly consistent and statistically significant (4.41±0.50 vs. 3.87±0.65, P<0.001). Figure 5 shows the end-diastolic short-axis images of Conv-cine (the first column), ACS-cine pre_MOCO (the second column), and ACS-cine MOCO (the third column) sequences. The third column of ACS-cine MOCO displayed smoother myocardial outline with no motion-related artifacts or image distortion, while the first and second columns showed the slight blur of trabeculations and endocardial border. Figure 6 shows a 3-year-old boy with giant right coronary artery (RCA) and left anterior descending (LAD) aneurysms, prior myocardial infarction, heart rate 85 bpm, and LVEF 40%. The image quality of ACS-cine MOCO was also of excellent quality. Inter-observer agreement for image quality scoring was almost perfect (weighted kappa =0.94).
Figure 4.
Bar stacking charts showing the distribution of image quality scores for ACS-cine and Conv-cine. 2.2% of patients in the ACS-cine group had a score of 3, which corresponds to the bottommost row in the figure. ACS-cine, artificial intelligence-assisted compressed sensing cine; Conv-cine, conventional cine.
Figure 5.
End-diastolic short-axis images of Conv-cine, ACS-cine (pre_MOCO) and ACS-cine (MOCO) sequences at the basal (A-C), mid-ventricular (D-F), and apical (G-I) levels. Magnification boxes (insets) highlight the endocardial border differences between sequences. ACS-cine, artificial intelligence-assisted compressed sensing cine; Conv-cine, conventional cine; MOCO, motion correction.
Figure 6.
End-diastolic short-axis images of Conv-cine, ACS-cine (pre_MOCO) and ACS-cine (MOCO) sequences at the basal (A-C), mid-ventricular (D-F), and apical (G-I) levels in a 3-year-old boy. The mild off-resonance banding artifact in (G-I) (arrow). Magnification boxes (insets) highlight the endocardial border differences between sequences. ACS-cine, artificial intelligence-assisted compressed sensing cine; Conv-cine, conventional cine; MOCO, motion correction.
Discussion
In cardiac cine scanning, ACS-cine adopts artificial neural network-assisted CS acceleration in both spatial and temporal dimensions, significantly reduces the amount of k-space data acquisition, achieving a higher acceleration factor. In addition to the high acceleration factor, another core advantage of ACS-cine is reduced breath-hold requirements. It can complete data acquisition with free breathing, significantly reducing the need for patient breath-holding. This advantage is particularly prominent in pediatric patients. Koechli et al. supported the use of ACS-cine sequences in pediatric patients in their 2024 study, especially in children who have difficulty cooperating with breath-holding (17). In this study, all participating children, even those with heart failure, were able to quickly complete cardiac function examinations using the ACS-cine sequence during free breathing with shorter scan time and good image quality.
We observed small but statistically significant systematic differences in LVEF and LVESV measurements between ACS-cine and Conv-cine. They may be due to the ACS-cine technique making certain sacrifices in spatial resolution (1.25×1.25 vs. 1.07×1.07 mm2), with undersampling potentially leading to endocardial border blurring or errors introduced during the iterative reconstruction process. Additionally, the high heart rate and respiratory rate in children can also exacerbate image artifacts and reduce measurement accuracy. Similar findings have been reported in previous studies (18). In studies on pediatric patients, it was found that CS-cine had greater spatial blurring with increased acceleration (19). For children with Kawasaki disease and heart failure, a higher heart rate may pose a challenge in accurately defining the endocardial contour, and minor differences in the endocardial contour may be the reason for the observed overestimation of ventricular volume by ACS-cine. Therefore, during ACS-cine examinations, if the image quality is found to be poor, a Conv-cine sequence can be added to correct the results (20).
The higher image quality scores for ACS-cine, despite its lower in-plane spatial resolution, can be attributed to the effective reduction of motion artifacts through AI-enabled online MOCO. While the iterative reconstruction process may cause some initial blurring of the endocardial border due to undersampling, ACS-cine’s MOCO algorithm specifically addresses respiratory motion, resulting in sharper endocardial border definition in the final reconstructed images. Mild off-resonance banding artifacts were occasionally observed in apical slices, particularly at 3-T field strength. These artifacts are inherent to bSSFP sequences and were not considered to significantly impact the diagnostic quality of the images. Yan et al. also found that ACS-cine technology is comparable to traditional Conv-cine in terms of image quality (9).
We acknowledge that breath-hold Conv-cine is the current clinical gold standard for cardiac volumetric assessment. Because breath-held Conv-cine was not routinely performed across all participants to maintain protocol consistency in pediatric patients, our study evaluates ACS-cine against free-breathing Conv-cine with signal averaging rather than the standard breath-held cine technique. Among all 56 initial patients, the technical success rate for ACS-cine and Conv-cine is 96.4% and 92.8%. The use of free-breathing Conv-cine with signal averaging (four averages) as the comparator in this study was specifically designed to evaluate ACS-cine under identical physiological conditions. Signal averaging in free-breathing Conv-cine can reduce random noise but does not fully compensate for respiratory motion, potentially resulting in some blurring of myocardial borders. This may also explain why ACS-cine with MOCO achieved higher image quality scores despite lower spatial resolution. Complete visual blinding could not be fully guaranteed due to inherent sequence appearance differences.
The observed mean difference in LVEF and LVESV between ACS-cine and Conv-cine is unlikely to alter clinical management decisions in most patients with Kawasaki disease (17), but for patients with borderline LVEF values (e.g., 55–60%), the small systematic difference should be considered to tell if myocardial dysfunction happened (13,21). While the mean bias in Figure 3 was small, the 95% LoA were relatively wide for ventricular volumes. The clinical acceptability of these limits should be considered: for group-level comparisons (e.g., research studies), the agreement is clinically acceptable; however, at the individual patient level, the wide LoA suggest that ACS-cine measurements may not be directly interchangeable with Conv-cine for all clinical decisions, particularly when precise individual measurements are critical. For longitudinal follow-up, we recommend using the same imaging technique consistently to minimize inter-technique variability.
The main limitation of this study is the small sample size (45 cases) and the focus on Kawasaki disease patients, and the number of patients with severe dysfunction is insufficient for a meaningful statistical analysis. The exclusion of two patients due to severe motion artifacts related to patient non-cooperation may introduce selection bias, and our findings may not be generalizable to children with severe behavioral challenges during CMR. We acknowledge that complete blinding may have been challenging due to inherent differences in spatial resolution and image texture between the two sequences. However, we believe any potential bias was minimized by the use of an objective 5-point Likert scale, automated contouring protocols, and the high inter-observer agreement observed (ICC: 0.882–0.991). Future studies can expand the sample size and include more types of pediatric cardiac disease to verify the general applicability of ACS-cine technology. Additionally, further optimization of image reconstruction algorithms can help improve image accuracy and measurement consistency.
Conclusions
ACS-cine technology demonstrates significant time efficiency advantages in CMR with satisfactory image quality, although there are statistically significant differences in the measurement of certain cardiac function parameters (LVEF and LVESV) compared to free-breathing Conv-cine in patients with relatively preserved function. These differences may be clinically acceptable for most patients but should be considered when precise individual measurements are critical. While agreement is clinically acceptable for group comparisons, caution and consistent sequence selection are recommended for individual longitudinal tracking due to volume LoA.
Supplementary
The article’s supplementary files as
Acknowledgments
None.
Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Children’s Hospital of Fudan University (No. IRB-2023-X12). Written informed consent was obtained from parents or legal guardians for all patients.
Footnotes
Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0641/rc
Funding: The work was supported by the National Natural Science Foundation of China (No. 82100524 to L.H.) and the Research Project of Children’s Hospital of Fudan University (No. EKYX202425 to X.H.).
Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0641/coif). The authors have no conflicts of interest to declare.
Data Sharing Statement
Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0641/dss
References
- 1. Kramer CM, Barkhausen J, Bucciarelli-Ducci C, et al. Standardized cardiovascular magnetic resonance imaging (CMR) protocols: 2020 update. J Cardiovasc Magn Reson 2020;22:17.. 10.1186/s12968-020-00607-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Curtis AD, Cheng HM. Primer and Historical Review on Rapid Cardiac CINE MRI. J Magn Reson Imaging 2022;55:373-88.. 10.1002/jmri.27436 [DOI] [PubMed] [Google Scholar]
- 3. Mayr A, Reiter G, Beitzke D. Cardiac magnetic resonance imaging : Trends and developments. Radiologe 2020;60:1142-52. [Article in German].. 10.1007/s00117-020-00766-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Altmann S, Halfmann MC, Abidoye I, et al. Compressed sensing acceleration of cardiac cine imaging allows reliable and reproducible assessment of volumetric and functional parameters of the left and right atrium. Eur Radiol 2021;31:7219-30.. 10.1007/s00330-021-07830-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Zucker EJ. Compact pediatric cardiac magnetic resonance imaging protocols. Pediatr Radiol 2023;53:1336-51.. 10.1007/s00247-022-05447-y [DOI] [PubMed] [Google Scholar]
- 6. Hsiao A, Lustig M, Alley MT, et al. Rapid pediatric cardiac assessment of flow and ventricular volume with compressed sensing parallel imaging volumetric cine phase-contrast MRI. AJR Am J Roentgenol 2012;198:W250-9.. 10.2214/AJR.11.6969 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Rau A, Soschynski M, Taron J, et al. Artificial intelligence and radiomics : Value in cardiac MRI. Radiologie (Heidelb) 2022;62:947-53. [Article in German].. 10.1007/s00117-022-01060-0 [DOI] [PubMed] [Google Scholar]
- 8. Ismail TF, Strugnell W, Coletti C, et al. Cardiac MR: From Theory to Practice. Front Cardiovasc Med 2022;9:826283.. 10.3389/fcvm.2022.826283 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Yan X, Luo Y, Chen X, et al. From Compressed-Sensing to Deep Learning MR: Comparative Biventricular Cardiac Function Analysis in a Patient Cohort. J Magn Reson Imaging 2024;59:1231-41.. 10.1002/jmri.28899 [DOI] [PubMed] [Google Scholar]
- 10. Qin C, Schlemper J, Caballero J, et al. Convolutional Recurrent Neural Networks for Dynamic MR Image Reconstruction. IEEE Trans Med Imaging 2019;38:280-90.. 10.1109/TMI.2018.2863670 [DOI] [PubMed] [Google Scholar]
- 11.Lyu J, Ding Y, Zhong J, et al. Toward single breath-hold whole-heart coverage compressed sensing MRI using VAriable spatial-temporal LAtin hypercube and echo-Sharing (VALAS). In: Proceedings of the 27th annual meeting of ISMRM. 2019. [Google Scholar]
- 12. Jone PN, Romanowicz J, Browne L, et al. Imaging Evaluation of Kawasaki Disease. Curr Cardiol Rep 2022;24:1487-94.. 10.1007/s11886-022-01768-4 [DOI] [PubMed] [Google Scholar]
- 13. McCrindle BW, Rowley AH, Newburger JW, et al. Diagnosis, Treatment, and Long-Term Management of Kawasaki Disease: A Scientific Statement for Health Professionals From the American Heart Association. Circulation 2017;135:e927-99.. 10.1161/CIR.0000000000000484 [DOI] [PubMed] [Google Scholar]
- 14. Lin L, Li Y, Wang J, et al. Free-breathing cardiac cine MRI with compressed sensing real-time imaging and retrospective motion correction: clinical feasibility and validation. Eur Radiol 2023;33:2289-300.. 10.1007/s00330-022-09210-7 [DOI] [PubMed] [Google Scholar]
- 15.Chen EZ, Chen X, Lyu J, et al. Real-time cardiac cine MRI with residual convolutional recurrent neural network. arXiv:2008.05044 [Preprint]. 2020. Available online: https://archive.ismrm.org/2019/4752.html.
- 16. Xie Y, Tao H, Li X, et al. Prospective Comparison of Standard and Deep Learning-reconstructed Turbo Spin-Echo MRI of the Shoulder. Radiology 2024;310:e231405.. 10.1148/radiol.231405 [DOI] [PubMed] [Google Scholar]
- 17. Koechli M, Callaghan FM, Burkhardt BEU, et al. Accelerated cardiac magnetic resonance imaging using deep learning for volumetric assessment in children. Pediatr Radiol 2024;54:1674-85.. 10.1007/s00247-024-05978-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Morales MA, Assana S, Cai X, et al. An inline deep learning based free-breathing ECG-free cine for exercise cardiovascular magnetic resonance. J Cardiovasc Magn Reson 2022;24:47.. 10.1186/s12968-022-00879-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Naresh NK, Malone L, Fujiwara T, et al. Use of compressed sensing to reduce scan time and breath-holding for cardiac cine balanced steady-state free precession magnetic resonance imaging in children and young adults. Pediatr Radiol 2021;51:1192-201.. 10.1007/s00247-020-04952-2 [DOI] [PubMed] [Google Scholar]
- 20. Zucker EJ, Sandino CM, Kino A, et al. Free-breathing Accelerated Cardiac MRI Using Deep Learning: Validation in Children and Young Adults. Radiology 2021;300:539-48.. 10.1148/radiol.2021202624 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Jone PN, Tremoulet A, Choueiter N, et al. Update on Diagnosis and Management of Kawasaki Disease: A Scientific Statement From the American Heart Association. Circulation 2024;150:e481-500.. 10.1161/CIR.0000000000001295 [DOI] [PubMed] [Google Scholar]