Introduction

Highly-accelerated real-time cine MRI is a rapid technique that is particularly useful for patients with arrhythmia, dyspnea, and/or pediatric patients who may not be able to follow breath-hold instructions. Our previous study demonstrated that radial k-space sampling and compressed sensing achieves 12 to 16-fold acceleration and relatively accurate left ventricular functional parameters in adults¹. Imaging pediatric patients is technically challenging because their hearts are smaller and beat much faster. We sought to develop and evaluate a 32-fold accelerated real-time free-breathing cine MRI using radial k-space sampling with view-sharing (VS) and k-space weighted image contrast (KWIC) filtering², for the purpose of achieving high spatio-temporal resolution and accurate biventricular volumetric parameters in pediatric patients.

Methods

Human Subjects & Pulse Sequence: We retrospectively identified 14 pediatric patients with congenital heart disease (CHD) (mean age = 9±2 years, 13/1 males/females) who underwent a standard breath-hold cine acquisition and real-time free-breathing cine using radial k-space sampling pattern with tiny golden angle sequence = 32.04° (N = 5) ¹. To achieve high temporal resolution for pediatric patients, we reconstructed each time frame using 6 radial spokes, which achieves 32-fold acceleration. Relevant image parameters were summarized in Table 1.
Image reconstruction: Figure1a illustrates three different image formation schemes: default CS with 6 radial spokes per frame; CS with VS 18 (9 before and after native) adjacent radial spokes, and CS with VS and KWIC filtering to exclude the center of k-space in VS radial spokes. To accelerate image reconstruction, coil compression was used to produce eight virtual coils ³, and gpuNUFFT was applied throughout the reconstruction pipeline ⁴. We used temporal total variation (TTV) and temporal PCA (TPCA) as the two orthogonal sparsifying transforms using empirically derived optimal regularization weights.
Image quality analysis: For quantitative evaluation, we calculated the blur metric ⁵ and bi-ventricular volumetric parameters (EDV, ESV, SV and EF). For blur metric calculation, we limited the analysis to the heart region by manually cropping the FOV. Cardiac contours were segmented using the automatic AI tools followed by manual correction in Circle CVI42. For qualitative evaluation, two clinical readers independently graded the images using a 5-point Likert scale (See Table 2). We tested for parameter normality using the Shapiro-Wilk test. We compared the blur metric calculated from the clinical standard cine, real-time cine with CS alone, CS with VS, and CS with VS and KWIC filtering using one-way ANOVA with Bonferroni correction as the post hoc test. We compared visual scores and biventricular volumetric parameters between clinical standard cine and real-time cine - the group with the smallest blur metric (i.e., winner) - using two-tailed, paired t-test (Wilcoxon signed-rank, if not normally distributed). In addition, linear-regression and Bland-Altman analyses were conducted on biventricular volumetric parameters to determine the coefficient of determination (R²) and levels of agreement. A p < 0.05 was considered statistically significant for each statistical test.

Results

According to the Shapiro-Wilk test, the blur metric and biventricular volumetric parameters were normally distributed (statistic: [0.1, 1]), whereas the visual scores were not normally distributed (p < 0.05). Figure 1b shows representative cine images: clinical standard, real-time with CS alone, CS with VS, and CS with VS and KWIC filtering. Among real-time reconstructions, incorporating VS+KWIC produced the best image quality as shown. As summarized in Table 2, VS with KWIC filtering significantly (p<0.05) reduced the blur metric (0.39±0.03) compared with VS alone (0.44±0.03) and default CS (0.51±0.03). Thus, we used CS with VS+KWIC for visual and biventricular volumetric parameters comparison with clinical standard cine. The median conspicuity of endocardial border, temporal fidelity of wall motion, and artifact scores were not significantly different, whereas the noise score was significantly different. Nonetheless, all scores were above the clinically acceptable (3.0) cut point. There were no significant differences in biventricular volumetric parameters (p>0.05) except for the RV EDV and RV SV (p<0.05). As shown in Figure 2, linear regression showed a strong correlation (R²>0.85) between clinical standard cine and real-time cine with VS and KWIC filtering for all categories. Bland-Altman analysis (Figure 3) showed the following results: LV ventricular volumetric parameters (absolute mean-difference ≤ 0.38 ml for EDV, ESV and SV, absolute mean-difference = 0.34% for EF), RV ventricular volumetric parameters (absolute mean-difference ≤ 4.81ml for EDV, ESV and SV, absolute mean-difference = 0.73% for EF).

Conclusion

32-fold accelerated real-time free-breathing cine MRI with radial k-space sampling, VS, and KWIC filtering produces good image quality and relatively accurate biventricular volumetric parameters in pediatric patients, which may eliminate the need for sedation/anesthesia.