Ultra-rapid, Free-breathing, Real-time Cardiac Cine MRI Using GRASP Amplified with View Sharing and KWIC Filtering

Abstract

Purpose

To achieve ultra-high temporal resolution (approximately 20 msec) in free-breathing, real-time cardiac cine MRI using golden-angle radial sparse parallel (GRASP) reconstruction amplified with view sharing (VS) and k-space–weighted image contrast (KWIC) filtering.

Materials and Methods

Fourteen pediatric patients with congenital heart disease (mean age [SD], 9 years ± 2; 13 male) and 10 adult patients with arrhythmia (mean age, 62 years ± 8; nine male) who underwent both standard breath-hold cine and free-breathing real-time cine using GRASP were retrospectively identified. To achieve high temporal resolution, each time frame was reconstructed using six radial spokes, corresponding to acceleration factors ranging from 24 to 32. To compensate for loss in spatial resolution resulting from over-regularization in GRASP, VS and KWIC filtering were incorporated. The blur metric, visual image quality scores, and biventricular parameters were compared between clinical and real-time cine images.

Results

In pediatric patients, the incorporation of VS and KWIC into GRASP (ie, GRASP + VS + KWIC) produced significantly (P < .05) sharper x-y-t (blur metric: 0.36 ± 0.03, 0.41 ± 0.03, 0.48 ± 0.03, respectively) and x-y-f (blur metric: 0.28 ± 0.02, 0.31 ± 0.03, 0.37 ± 0.03, respectively) component images compared with GRASP + VS and conventional GRASP. Only the noise score differed significantly between GRASP + VS + KWIC and clinical cine; all visual scores were above the clinically acceptable (3.0) cutoff point. Biventricular volumetric parameters strongly correlated (R² > 0.85) between clinical and real-time cine images reconstructed with GRASP + VS + KWIC and were in good agreement (relative error < 6% for all parameters). In adult patients, the visual scores of all categories were significantly lower (P < .05) for clinical cine compared with real-time cine with GRASP + VS + KWIC, except for noise (P = .08).

Conclusion

Incorporating VS and KWIC filtering into GRASP reconstruction enables ultra-high temporal resolution (approximately 20 msec) without significant loss in spatial resolution.

Keywords: Cine, View Sharing, k-Space–weighted Image Contrast Filtering, Radial k-Space, Pediatrics, Arrhythmia, GRASP, Compressed Sensing, Real-Time, Free-Breathing

Supplemental material is available for this article.

© RSNA, 2024

Summary

This study demonstrates that the incorporation of view sharing and k-space–weighted image contrast filtering into golden-angle radial sparse parallel reconstruction achieves ultra-high temporal resolution (approximately 20 msec) in free-breathing real-time cine MRI.

Key Points

• ■ Incorporation of view sharing (VS) and k-space–weighted image contrast (KWIC) filtering into golden-angle radial sparse parallel (GRASP) reconstruction (ie, GRASP + VS + KWIC) achieved a good balance between high temporal resolution (approximately 20 msec) and spatial resolution in free-breathing real-time cine MRI.

• ■ Real-time cine reconstructed with GRASP + VS + KWIC produced significantly sharper x-y-t and x-y-f component images (blur metric = 0.36 ± 0.03 and 0.28 ± 0.02) compared with GRASP + VS (blur metric = 0.41 ± 0.03 and 0.31 ± 0.03) and conventional GRASP (blur metric = 0.48 ± 0.03 and 0.37 ± 0.03).

• ■ Real-time cine reconstructed with GRASP + VS + KWIC produced similar image quality and biventricular volumetric parameters in pediatric patients with congenital heart disease as clinical breath-hold cine and better image quality in the adult patients with arrhythmia.

Introduction

Cardiac function evaluation with imaging plays an important role in both diagnosis and therapy monitoring of patients with heart disease (1). Cardiac MRI has shown to be an excellent tool for imaging left ventricular volumes and function and is considered the reference standard for this evaluation (2). Although electrocardiographically (ECG) gated, breath-hold cine MRI with balanced steady-state free precession readout is the clinical standard (3), it has two noted disadvantages: lengthy imaging time and sensitivity to irregular heart rhythm and breathing motion. In such clinical scenarios (eg, atrial fibrillation, dyspnea, and/or pediatric patients who may not be able to follow breath-hold instructions), this technique may produce poor or nondiagnostic image quality.

One approach to overcoming these technical challenges is the use of free-breathing, real-time cine MRI (4,5), which has several advantages mirroring echocardiography (6). Previous compressed sensing studies describing accelerated real-time cine pulse sequences report temporal resolutions ranging from 30 to 47 msec using various techniques, such as Cartesian k-space sampling using compressed sensing (7–9) and non-Cartesian sampling with compressed sensing (5,10). For pediatric patients with rapid heart rates (>85 beats per minute) and adult patients with arrhythmia (eg, atrial fibrillation) or tachycardia, it may be necessary to achieve even higher temporal resolution; for example, standard two-dimensional (2D) echocardiography achieves a temporal resolution of 25 msec (11). Such target acceleration necessitates additional techniques beyond conventional golden-angle radial sparse parallel (GRASP) reconstruction (12) (ie, radial compressed sensing with parallel imaging) because over-regularization in GRASP may lead to substantial image blurring.

One approach to achieving high temporal resolution without substantial loss in spatial resolution is view sharing (VS) (13). Unlike Cartesian k-space sampling, radial k-space sampling is better suited for VS because every radial spoke samples the center of k-space (14); in addition, using golden angles (such as the fifth and seventh Fibonacci sequences of golden angles used in this study) (15) enables retrospective binning to tailor temporal resolution based on the patient’s heart rate. However, radial k-space sampling is also sensitive to motion blurring from VS because the center of k-space is sampled by every radial spoke. Inspired by the vastly undersampled isotropic projection reconstruction technique (16), we sought to incorporate k-space–weighted image contrast (KWIC) filtering (17) to filter out the center of shared k-space lines in GRASP reconstruction. The purpose of this study was to achieve ultra-high temporal resolution (approximately 20 msec) for free-breathing, real-time cine MRI and evaluate its performance in pediatric patients with congenital heart disease (CHD) and adult patients with arrhythmia.

Materials and Methods

Study Sample

We retrospectively identified 14 pediatric patients with CHD (mean age [SD], 9 years ± 2; 13 male and one female; November 2018–January 2022) and 10 adult patients with arrhythmia (mean age, 62 years ± 8; nine male and one female; coefficient of variation in R-R intervals > 10%; April 2016–December 2022) who underwent both standard breath-hold cine MRI and free-breathing real-time cine (5) from multiple studies conducted under separate institutional review boards; none of the data have been previously reported. Figure 1A illustrates the workflow for free-breathing real-time cine data acquisition. See Tables S1 and S2 for relevant image parameters and Tables S3 and S4 for the relevant clinical profiles of our pediatric and adult patient samples, respectively. In the pediatric patients, all real-time cine data were acquired before sedation, and five of the 14 clinical cine datasets were obtained during sedation. All patients with CHD had undergone surgery, and cardiac MRI was conducted as part of routine follow-up. All patients (or guardians of the pediatric patients) provided written informed consent and agreed to future analysis of data. This study was performed in accordance with protocols approved by our institutional review board and was Health Insurance Portability and Accountability Act compliant.

Figure 1:

Workflows for data acquisition and image reconstruction. (A) The pulse sequence diagram outlines the sampling strategy. Dummy scans (green arrows) were prescanned during the first heartbeat to approach steady-state magnetization. Subsequently, free-breathing real-time radial data were acquired in the subsequent four heartbeats using the tiny golden angle (TGA) sequence (black arrows). Multiple sections were consecutively acquired until the whole heart was covered. (B) Reconstruction process for free-breathing real-time radial cine imaging with the incorporation of view sharing (VS) and k-space–weighted image contrast (KWIC) filtering techniques. ECG = electrocardiography, GRASP = golden-radial angle sparse parallel, NUFFT = nonuniform fast-Fourier transform, PCA = principal component analysis.

MRI Hardware

MRI was performed with two 1.5-T whole-body MRI scanners (adults were scanned with MAGNETOM Avanto and Aera [Siemens Healthineers], and children were scanned with Aera), equipped with a gradient system capable of achieving a maximum strength of 45 mT/m and maximum slew rate of 200 T/m/sec. The body coil was used for radiofrequency excitation, and both body matrix and spine coil arrays (approximately 15–18 elements for Avanto and approximately 30 elements for Aera) were used for signal reception.

VS and KWIC Filtering

To achieve high temporal resolution equivalent to that seen with 2D echocardiography (approximately 25 msec), we reconstructed our proposed free-breathing real-time cine data using six radial spokes per time frame (temporal resolution = echo spacing × radial spokes per time frame = approximately 3 msec × 6 = approximately 18 msec) (Fig 2A). To minimize the loss of spatial resolution, we incorporated VS (Fig 2B); to minimize the loss of temporal resolution (ie, minimize temporal blurring), we incorporated KWIC filtering (Fig 2C), similar to the “tornado” temporal filter used in the vastly undersampled isotropic projection reconstruction technique (16,18). Appendix S1 provides more detail on the three image reconstruction schemes and the optimization of the number of VS spokes and the width of the KWIC filter.

Figure 2:

A schematic illustrates how view sharing (VS) and k-space–weighted image contrast (KWIC) filtering are incorporated into a golden-radial angle sparse parallel (GRASP) framework to improve the image quality by leveraging the high frequency k-space data from adjacent radial spokes. (A) Conventional GRASP reconstruction: Six radial spokes were used to reconstruct each time frame image to achieve high temporal resolution. (B) GRASP + VS reconstruction: 24 rays (VS 18 adjacent radial spokes [nine before and nine after]) were used to reconstruct each time frame image to improve image quality. (C) GRASP + VS + KWIC reconstruction: k-space center of VS radial spokes was excluded to suppress motion blurring from shared k-space lines.

Image Reconstruction

Image reconstruction was performed offline on a graphics processing unit workstation (Tesla V100, 16 GB memory [NVIDIA]; 32 Xeon E5–2620 v4, 128 GB memory [Intel]) running on a Linux operating system (Ubuntu1 8.04). Custom-made codes were written with MATLAB, R2020b (MathWorks). Figure 1B provides a flowchart of image reconstruction. For more details, please refer to Appendix S1. The reconstruction time was approximately 10 minutes per 2D plane with 150 frames.

Quantitative Analysis of Image Quality

To quantitatively evaluate the image quality of clinical cine and free-breathing real-time cine images reconstructed using three different schemes (conventional GRASP, GRASP + VS, and GRASP + VS + KWIC), we calculated the blur metric (19) (ranging from 0 [sharp] to 1 [blur]) on both the x-y-t (ie, spatial resolution) and x-y-f (ie, temporal resolution) component images in Matlab (see Fig 3). Appendix S1 provides more details. The reconstruction scheme generating the smallest blur metric value (ie, sharpest) of the real-time cine was used for further image analysis.

Figure 3:

Representative x-y-f images show five frequency components: −2 from direct current (DC), −1 from DC, DC (frequency = 0 Hz), +1 from DC, and +2 from DC. The blur metric was calculated on the ± first component from DC; the mean blur metric was 0.39 for golden-radial angle sparse parallel (GRASP), 0.32 for GRASP + view sharing (VS), and 0.28 for GRASP + VS + k-space–weighted image contrast (KWIC).

Visual Analysis of Image Quality

Image sets were randomized and de-identified for independent evaluation by two raters (C.K.R., J.D.R. for the pediatric data, D.C.L., B.D.A. for the adult data) using a five-point Likert scale for the following four categories: conspicuity of endocardial border (ie, spatial resolution and image contrast) and temporal fidelity (1 = nondiagnostic, 2 = poor, 3 = clinically acceptable, 4 = good, 5 = excellent) and noise and artifact levels (1 = nondiagnostic, 2 = severe, 3 = moderate, 4 = mild, 5 = minimal) (see Appendix S1 for additional information).

Cardiac Functional Parameters

Biventricular functional parameters were calculated by analyzing both the clinical breath-held cine and free-breathing real-time cine images with the highest image quality using Circle CVI42, version 5.13.10 (Cardiovascular Imaging). For more details, refer to Appendix S1.

Statistical Analysis

Appropriate statistical analyses were conducted using Matlab, R2020b. A P value less than .05 was considered to indicate a statistically significant difference for each statistical test. Appendix S1 supplies more information on statistical analyses.

Results

According to the Shapiro–Wilk test, the blur metric and volumetric parameters were normally distributed (statistic: [0.1, 1]), whereas the visual scores were not normally distributed (P < .05).

Comparison of Clinical Cine and Free-breathing Real-time Cine in Pediatric Patients

Among free-breathing real-time cine reconstructions, incorporating VS + KWIC into GRASP produced the sharpest image quality both spatially and temporally. VS with KWIC filtering significantly (P < .05) reduced spatial blurring and temporal blurring (blur metric: 0.36 ± 0.03 and 0.28 ± 0.02, respectively) compared with VS alone (blur metric: 0.41 ± 0.03 and 0.31 ± 0.03) and conventional GRASP (blur metric: 0.48 ± 0.03 and 0.37 ± 0.03) (Table 1). Thus, the remaining real-time cine results correspond to GRASP + VS + KWIC. In addition, the blur metric on the x-y-t (ie, spatial) and x-y-f component (ie, temporal) images of the free-breathing real-time cine reconstructed with GRASP + VS + KWIC (0.36 ± 0.03 and 0.28 ± 0.02) was significantly smaller than that of the clinical cine (0.42 ± 0.04 and 0.36 ± 0.03).

Table 1:

Blur Metric Parameters Obtained from Pediatric Patients

For visual analysis (Table 2), no evidence showed a difference in any score (P > .10), except the noise score (P < .05), and all scores were above the clinically acceptable (3.0) cutoff point. The interrater reliability in visual scores for the four categories was slight to moderate (κ = [0.13, 0.43]) for clinical cine and poor to slight (κ = [0, 0.13]) for real-time cine. Figure 4 shows the ECG trace extracted from the raw data file of a pediatric patient and the corresponding cine image sets, where GRASP + VS + KWIC produced the highest real-time cine image quality. See Movie 1 for dynamic display.

Table 2:

Visual Scores Obtained from Pediatric Patients

Figure 4:

(A) Electrocardiographic (ECG) trace extracted from the raw data file, where the green box indicates data acquisition per section. (B) Representative multisection cine images from a 12-year-old male patient (heart rate, 75 beats per minute; congenital heart disease type: transposition of the great arteries). Real-time cine with golden-radial angle sparse parallel (GRASP) + view sharing (VS) + k-space–weighted image contrast (KWIC) produced the best image quality that approximated those produced with clinical cine MRI. For the corresponding movie file, see Movie 1.

Movie 1:

Download video file ^{(1.7MB, mp4)}

Movie display of representative pediatric cine images as shown in Figure 4B: standard cine (row 1), free-breathing real-time cine reconstructed with conventional GRASP (row 2), GRASP+VS (row 3) and GRASP+VS+KWIC (row 4). To combine the standard cine and real-time cine movies into one video for convenient display, we interpolated the standard cine through time to match with the real-time cine time frames. Display rate = 1s/temporal resolution = 51 fps. VS = view-sharing, KWIC = k-space weighted image contrast, GRASP = golden-radial angle sparse parallel, fps = frames per second.

Summarizing the results of all pediatric patients, we found no evidence of differences in biventricular volumetric parameters (P > .05), except for the right ventricular end-diastolic volume and right ventricular stroke volume (P < .05) (Table 3). As shown in Figure 5, volumetric parameters were strongly correlated (R² > 0.85) and in excellent agreement (relative error < 6%).

Table 3:

Biventricular Volumetric Parameters Obtained from Pediatric Patients

Figure 5:

(A) Linear regression plots compare clinical standard cine (reference) and free-breathing real-time cine MRI reconstructed with golden-radial angle sparse parallel (GRASP) + view sharing (VS) + k-space–weighted image contrast (KWIC) filtering. A strong correlation (R² > 0.85) was observed for all parameters. Solid line represents the fitted line; dotted lines represent 95% CIs. (B) Bland–Altman plots show agreement in biventricular parameters between clinical standard cine and free-breathing real-time cine MRI reconstructed with GRASP + VS + KWIC filtering: relative error less than 6% for all parameters. EDV = end-diastolic volume, EF = ejection fraction, ESV = end-systolic volume, LV = left ventricle, RV = right ventricle, SV = stroke volume.

Comparison of Clinical Cine and Free-breathing Real-time Cine in Adult Patients

As summarized in Table 4, all visual scores were significantly lower (P < .05) for clinical cine compared with the real-time cine reconstructed with GRASP + VS + KWIC, whereas no evidence showed a difference in noise scores (P = .08). The interrater reliability in visual scores for the four categories was fair to substantial (κ = [0.24, 0.61]) for clinical cine and poor to substantial (κ = [−0.09, 0.63]) for real-time cine. Figure 6 shows the ECG trace extracted from the raw data file of an adult patient with arrhythmia (coefficient of variation in R-R interval = 28%) and the corresponding image sets, where real-time cine produced better image quality than clinical cine. See Movie 2 for dynamic display.

Table 4:

Summary of Visual Scores Obtained from the Adult Group

Figure 6:

(A) Electrocardiographic (ECG) trace extracted from the raw data file, where the green box indicates data acquisition per section. (B) Representative multisection cine images from a 65-year-old male patient with arrhythmia (coefficient of variation in R-R interval = 28%). Real-time cine with golden-radial angle sparse parallel (GRASP) + view sharing (VS) + k-space–weighted image contrast (KWIC) produced better image quality compared with those produced with clinical cine MRI. For the corresponding movie file, see Movie 2.

Movie 2:

Download video file ^{(2.4MB, mp4)}

Movie display of cine images from the representative adult with arrhythmia (CV R-R = 28%) as shown in Figure 6B: standard cine (row 1), free-breathing real-time cine reconstructed with GRASP+VS+KWIC (row 2). To combine the standard cine and real-time cine movies into one video for convenient display, we interpolated the standard cine through time to match with the real-time cine time frames. Display rate = 1s/temporal resolution = 60 fps. CV = coefficient of variation, VS = view-sharing, KWIC = k-space weighted image contrast, GRASP = golden-radial angle sparse parallel, fps = frames per second.

Movie 3:

Download video file ^{(850.5KB, mp4)}

Movie display of cine images shown in Figure S3. Display rate = 1s/temporal resolution = 51 fps. fps = frames per second, TTV = temporal total variation.

Discussion

This study describes development and evaluation of real-time cine MRI with ultra-high temporal resolution (approximately 20 msec) for challenging clinical scenarios, such as pediatric patients with limited cooperation and adult patients with arrhythmia. The incorporation of VS and KWIC filtering into the GRASP reconstruction pipeline enabled a good balance between high temporal resolution and spatial resolution while producing high image quality and relatively accurate biventricular functional parameters.

Our study compares favorably with similar studies. First, Levine et al (20) demonstrated VS with compressed sensing in Cartesian k-space sampling; in this technique, the sampling scheme alternates between densely sampled center and sparsely sampled edges of k-space. The effective temporal resolution is, therefore, prefixed by the time spent on sampling edges of k-space that are shared across reconstructed frames. In our strategy using GRASP with VS + KWIC, temporal resolution can be arbitrarily defined through rebinning. This provides an advantage because it provides a means to retrospectively define an optimal temporal resolution tailored to the patient’s heart rate. Second, Zhang et al (14) demonstrated VS with sliding window in radial k-space sampling; in this technique, however, VS introduces motion blurring from shared k-space lines. In our strategy with KWIC filtering, we suppressed motion blurring by filtering out the center of shared k-space lines. Third, Uecker et al (21) demonstrated spatial resolution (2 × 2 mm²) and temporal resolution (18 msec) similar to ours. However, their reconstruction pipeline involved the application of both temporal median and spatial filters during postprocessing, inevitably leading to a certain degree of image blurring.

Our evaluation in a limited number of pediatric patients and adult patients with arrhythmia suggests that the proposed real-time cine with VS and KWIC produces diagnostically acceptable image quality and relatively accurate biventricular functional parameters in pediatric patients, as well as high image quality in adult patients with arrhythmia. The statistically significant differences in right ventricular end-diastolic volume and right ventricular stroke volume between clinical cine and real-time cine were not surprising because the differences in spatial resolution may be more pronounced for the thinner right ventricle than the left ventricle. The relative difference in right ventricular end-diastolic volume was within 5% and in right ventricular stroke volume was slightly lower than 6%.

This study had several limitations. First, it did not evaluate the clinical effect of improved temporal resolution for diagnosing heart disease in pediatric patients or adult patients with arrhythmia. A future study should evaluate the clinical utility of our real-time cine in a large cohort of pediatric patients or adult patients with arrhythmia. Second, the image reconstruction was performed offline and took approximately 10 minutes per 2D plane with 150 frames. Therefore, the term real-time in this context pertains to the fact that acquisition does not require repetition (22). For real-time display, it may be necessary to implement deep learning solutions (23–25) and integrate them into a vendor reconstruction pipeline (eg, Siemens Framework for Image Reconstruction Environments [FIRE] framework [26]) to enable inline reconstruction. Third, the stack of short-axis planes analyzed for ventricular functional parameters included a mixture of different respiratory states, suggestive of confounding effects by respiration. Correcting for this confounding effect was beyond the scope of this study.

In conclusion, the incorporation of VS and KWIC filtering into GRASP reconstruction achieves ultra-high temporal resolution in free-breathing real-time cine MRI. Our proposed real-time cine MRI may be a promising solution for challenging clinical scenarios, such as pediatric patients with CHD and adult patients with arrhythmia.

Data Availability Statement

Our custom-made code written in MATLAB for reconstructing real-time cine image is made available in GitHub (https://github.com/Lexiaozi-Fan/rtCine_VS_KWIC). The de-identified data will be made available once a data sharing agreement is established upon request.