60-S Retrogated Compressed Sensing 2D Cine of the Heart: Sharper Borders and Accurate Quantification

Background and objective: Real-time compressed sensing cine (CSrt) provides reliable quantification for both ventricles but may alter image quality. The aim of this study was to assess image quality and the accuracy of left (LV) and right ventricular (RV) volumes, ejection fraction and mass quantifications based on a retrogated segmented compressed sensing 2D cine sequence (CSrg). Methods: Thirty patients were enrolled. Each patient underwent the reference retrogated segmented steady-state free precession cine sequence (SSFPref), the real-time CSrt cine and the segmented retrogated prototype CSrg sequence providing the same slices. Functional parameters quantification and image quality rating were performed on SSFPref and CSrg images sets. The edge sharpness, which is an estimate of the edge spread function, was assessed for the three sequences. Results: The mean scan time was: SSFPref = 485.4 ± 83.3 (SD) s (95% CI: 454.3–516.5) and CSrg = 58.3 ± 15.1 (SD) s (95% CI: 53.7–64.2) (p < 0.0001). CSrg subjective image quality score (median: 4; range: 2–4) was higher than the one provided by CSrt (median: 3; range: 2–4; p = 0.0008) and not different from SSFPref overall quality score (median: 4; range: 2–4; p = 0.31). CSrg provided similar LV and RV functional parameters to those assessed with SSFPref (p > 0.05). Edge sharpness was significantly better with CSrg (0.083 ± 0.013 (SD) pixel−1; 95% CI: 0.078–0.087) than with CSrt (0.070 ± 0.011 (SD) pixel−1; 95% CI: 0.066–0.074; p = 0.0004) and not different from the reference technique (0.075 ± 0.016 (SD) pixel−1; 95% CI: 0.069–0.081; p = 0.0516). Conclusions: CSrg cine provides in one minute an accurate quantification of LV and RV functional parameters without compromising subjective and objective image quality.


Introduction
Cardiac magnetic resonance (CMR) is the reference standard method for quantification of volumes, ejection fraction (EF) and mass of left (LV) and right ventricles (RV) [1][2][3]. Reliable volumes assessment is required since EF has a strong prognostic value regarding clinical outcomes and survival [4][5][6]. However, besides steady-state free precession cine images essential for quantification, phase contrast angiography, gadolinium enhanced imaging, and additional sequences may be recommended depending on heart conditions,

Study Population
From March to April 2019, 30 consecutive adult patients referred for rest CMR were included. Exclusion criteria were grown-up congenital heart disease work-up or follow-up, underaged patients, patients suffering from arrhythmia for whom the use of prospective ECG gating was necessary, MRI contraindications and patient refusal. Patients gave informed consent, and the protocol was approved by our Institutional Ethics Committee.

Imaging Protocol
CMR studies were performed on a 1.5-T scanner (MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany). Every patient underwent three series of cine images: first the reference retrogated segmented multi-breath-hold SSFP sequence (SSFP ref ); then the CS-accelerated SSFP real-time sequence (CS rt ) acquired in two breath-holds, and finally the retrogated segmented SSFP prototype with CS-fashioned acceleration requiring three breath-holds (CS rg ). One LV 2-chamber, one RV 2-chamber, one 4-chamber slice and a LV short-axis stack covering the entire ventricles were acquired with the three above-cited sequences, providing identical slice position, thickness, and number. Imaging parameters for the three sequences are summarized in Table 1. Table 1. Imaging parameters of the reference steady-state free precession cine, real-time compressed sensing cine and segmented retrogated compressed sensing cine.  15.0 ± 1.2 2 a 3 a Cycles per slice-n 8 a 1 a 2 a Cycles of iterative reconstruction-n - 40 40 Data are expressed as mean ± standard deviation in the absence of any indication. a Constant value. Abbreviations: SSFP ref , reference steady-state free precession cine; CS rt , real-time compressed sensing cine; CS rg , segmented retrogated compressed sensing cine.

Cine Images Quality Assessment
A 3-step evaluation was performed for the three sequences. First, a subjective overall image quality was assessed using a 4-point Likert scale (1: non diagnostic, 2: fair, 3: good, 4: excellent).
Then acquisition quality was evaluated using a standardized score based on the "LVfunction cine SSFP" section of the criteria established from the European CMR registry, evaluating the artifact detection [15] (p. 3). This score was modified, removing its four last items since their score was systematically null, accordingly to our center practice (Table 2). This score increased with acquisition impairment.

Score 21
Modified score (items 1 to 8) 10 The four last items were nulled since acquisitions were repeated when orientation was not appropriated (item 9 = 0); all acquisitions were performed using a 8-mm thickness (item 10 score = 0) and a 2-mm gap (item 11 score = 0), horizontal and vertical long axis views were systematically acquired (item 12 score = 0). Consequently, italic criteria were not applied, and only bold criteria were used for objective quality assessment in our study, providing a maximum score of 10 points. The more artifacts there were, the higher the score was. Abbreviations: LV, left ventricle; SSFP, steady-state free precession.
Finally, the edge sharpness (ε) between myocardium and LV blood pool was measured on end-diastole 4-chamber view ( Figure 1). This assessment was performed using a MATLAB (version R2015a, The MathWorks, Natick, MA, USA) homemade script. An intensity profile line was drawn perpendicularly to the mid-cavity interventricular septum border with the LV blood pool at end-diastole [31][32][33]. The ε value was a spatial frequency (pixel −1 ) calculated as the inverse of the distance separating the two points corresponding to 20% and 80% of the difference between the minimum and maximum intensities along this line. The four last items were nulled since acquisitions were repeated when orientation was not appropriated (item 9 = 0); all acquisitions were performed using a 8-mm thickness (item 10 score = 0) and a 2-mm gap (item 11 score = 0), horizontal and vertical long axis views were systematically acquired (item 12 score = 0). Consequently, italic criteria were not applied, and only bold criteria were used for objective quality assessment in our study, providing a maximum score of 10 points. The more artifacts there were, the higher the score was. Abbreviations: LV, left ventricle; SSFP, steady-state free precession.
Finally, the edge sharpness (ε) between myocardium and LV blood pool was measured on end-diastole 4-chamber view ( Figure 1). This assessment was performed using a MATLAB (version R2015a, The MathWorks, Natick, MA, USA) homemade script. An intensity profile line was drawn perpendicularly to the mid-cavity interventricular septum border with the LV blood pool at end-diastole [31][32][33]. The ε value was a spatial frequency (pixel −1 ) calculated as the inverse of the distance separating the two points corresponding to 20% and 80% of the difference between the minimum and maximum intensities along this line.  The edge sharpness was calculated as the inverse of the distance d (in pixels) from the positions corresponding to 20% and 80% (red stars) of the difference between the maximum and minimum signal intensities (black crosses) along the profile line and was expressed in pixel −1 . Abbreviations: SSFP ref , reference steady-state free precession; CS rt , real-time compressed sensing; CS rg , retrogated compressed sensing; ε, edge sharpness; Ymax, maximum signal intensity; Ymin, minimum signal intensity; d, distance along the intensity (pixels).

Functional Evaluation
Assessment of end-diastolic volumes (EDV), end-systolic volumes (ESV), stroke volumes (SV) and EF were performed for both ventricles as well as LV mass (LVM). These parameters were measured on short-axis stacks with semi-automated segmentation with manual correction of the LV endocardium and epicardium while manual segmentation of the RV endocardium was necessary, using dedicated 4D analysis software (Cardiac MR analysis workflow, Syngo.via VB30A, Siemens Healthcare, Erlangen, Germany). Fourchamber, LV and RV 2-chamber slices were used to define mitral and tricuspid valve planes to ensure optimal assessment of ventricular bases.

Conditions of Image Analysis
The three datasets were independently analyzed by a 4-year experience CMR radiologist (LG). After anonymization and randomization performed for each sequence, each dataset was analyzed separately. Each analysis session, which evaluated all the images of the same sequence, was separated from the previous one by one month. A second radiologist with 11 years of experience in CMR (BL) segmented both ventricles on CS rg images for interrater variability assessment.

Statistics Analysis
Categorical data were presented as numbers (percentage) and continuous variables as mean ± standard deviation (SD) (95% confidence interval (CI)) in the case of normal distribution or median (range: minimum-maximum) in other cases. Variable normality was assessed using the D'Agostino-Pearson test.
Paired Wilcoxon signed-rank test was used to compare subjective image qualities and acquisition qualities between CS rg and SSFP ref or CS rt . An analysis of variance was performed to compare edge sharpness and acquisition times of the three sequences. CS rg and SSFP ref mean functional parameters were compared using a Student's t test, with linear regression and Bland-Altman analysis to assess the agreements between both methods. Inter and intra-observer variabilities were assessed using intra-class coefficient correlation. Significance of the test was defined by values of p < 0.05.
Regarding the acquisition quality based on the EuroCMR registry, the CSrg sequence

Volumes, Functions and Mass Quantification
Good agreements were yielded by Bland-Altman and linear regression analyses for both LV (Figure 4) and RV ( Figure 5) assessments. No significant difference was demonstrated regarding LVM, LV and RV volumes (EDV, ESV, SV) and EF (p > 0.05) ( Table 4). Intrarater variability was excellent, demonstrating intraclass correlation coefficients (ICC) greater than 0.99 for both ventricles, as were interrater variabilities for LV (ICC ≥ 0.97) and RV (ICC ≥ 0.96).

Volumes, Functions and Mass Quantification
Good agreements were yielded by Bland-Altman and linear regression analyses for both LV (Figure 4) and RV ( Figure 5) assessments. No significant difference was demonstrated regarding LVM, LV and RV volumes (EDV, ESV, SV) and EF (p > 0.05) ( Table 4). Intrarater variability was excellent, demonstrating intraclass correlation coefficients (ICC) greater than 0.99 for both ventricles, as were interrater variabilities for LV (ICC ≥ 0.97) and RV (ICC ≥ 0.96).

Volumes, Functions and Mass Quantification
Good agreements were yielded by Bland-Altman and linear regression analyses for both LV (Figure 4) and RV ( Figure 5) assessments. No significant difference was demonstrated regarding LVM, LV and RV volumes (EDV, ESV, SV) and EF (p > 0.05) ( Table 4). Intrarater variability was excellent, demonstrating intraclass correlation coefficients (ICC) greater than 0.99 for both ventricles, as were interrater variabilities for LV (ICC ≥ 0.97) and RV (ICC ≥ 0.96).

Discussion
Our clinical study performed a comprehensive evaluation of a retrogated CS sequence in daily practice. Results are in line with preliminary CS rg tests on eight healthy volunteers reported by Forman et al., regarding LVEF, LVEDV and LVESV, which were similar to SSFP ref quantification [34]. However, the study population was small and LVM, RV volumes and EF were not assessed. In the present study, no significant difference was demonstrated regarding LVM, LV and RV volumes (EDV, ESV and SV) and EF.
First-generation 49-ms temporal resolution CS rt sequence provided LVEDV underestimation and LVM overestimation, which were clinically insignificant or smaller than intra or interrater variabilities [17,22,25,26,35]. Moreover, LVEDV underestimation was also reported with other acceleration techniques such as radial gradient-echo or k-space parallel imaging [36,37]. These differences were not depicted with CS rg . This might be explained by the improved edge sharpness facilitating segmentation, retrospective ECG gating allowing acquisition of the last phases of the cardiac cycle and the better temporal resolution (37 ms) [38]. This observation is in line with previous studies suggesting an optimal temporal resolution for accurate steady-state free precession quantification below 45 ms [29]. Since quantification is a major CMR point of interest, the absence of significant difference regarding each LV and RV parameter seems promising for clinical implementation [1]. Moreover, the high intra and interrater reproducibility allows the use of CS rg for chronic status follow-up such as heart failure, anthracycline induced cardiotoxicity and other cardiomyopathies [39][40][41].
The overall image quality alteration using CS rt has already been assessed in a previous publication [22]. The higher score provided by CS rg confirmed the image quality improvement. Indeed, CS rt interpolation was responsible for smoothed images, which required physicians to get used to this rendering. The absence of difference between acquisitions qualities (EuroCMR quality score) was expected. Indeed, no difference had been demonstrated regarding the first-generation CS sequence. The evaluated prototype features partial Fourier switch-off and segmented acquisition, which were not supposed to generate more artifacts. Regarding edge sharpness, the absence of interpolation and acquisition of more data improves boarder delineation. Not only was ε CSrg better than ε CSrg but it was similar to ε SSFPref . Edge sharpness was chosen as a metric for intrinsic image quality assessment due to its simple and reproducible implementation. Moreover, edge sharpness is an estimate of the edge spread function whose derivative is a line-spread function [33]. Fast Fourier transform of the latter gives the task-based modulation transfer function (MTF Task ); edge sharpness can be considered as a reasonable approximation of MTF Task , which is widely used to evaluate the spatial frequency response of an imaging system, even using iterative reconstructions [42].
Predictably, CS rg scan time was approximatively twice as long as the first-generation CS sequence since acquisitions were segmented on two heart beats in contrast to real-time imaging. However, as compared to SSFP ref , the mean CS rg scan time of about 60 s still provides an 8.7-fold acceleration factor. The study protocol divided CS rg acquisition in 3 stacks to shorten breath-holdings. Nevertheless, this setting is adjustable, such as the number of heart beats required for one slice to reduce apneas or increase spatiotemporal resolution. We chose to use CS to improve scan time, with workflow fluency being a major point of concern. However, it is possible to take advantage of this acceleration to improve spatial resolution, or maybe more interestingly, temporal resolution. Indeed, in the field of feature tracking in CMR, a resolution at least over 30 frames per cycle is recommended for accurate strain assessment, which is more time-consuming when using SSFP ref [43]. Such settings may facilitate further feature tracking studies, though the impact of CS based reconstructions should be evaluated regarding strain analysis reliability.
The image quality improvement provided by the prototype CS rg as compared to the first-generation real-time CS rt should facilitate the spread of CS use in daily practice. However, the 2-shot acquisition is responsible for the loss of the real-time acquisition. Futures generations of CS cine sequences, maintaining the CS rt real-time feature and the CS rg image quality, should be oriented towards the application of additional motion correction algorithms to provide free-breathing acquisition with preserved image quality.

Limitations
The size of the population was limited and cannot represent the whole variety of cardiac conditions encountered in daily practice. Nevertheless, the main objective of this study was to assess the image quality recovery as compared to SSFP ref and CS rt . To facilitate the standardized assessment of edge sharpness, we chose to exclude congenital heart disease anatomy. Among the cardiac conditions assessed by CMR, dilated cardiomyopathy is a frequent pathology, usually responsible for shortness of breath, in which the LV wall may be thinned. The decreased wall thickness may be challenging for endocardium and epicardium delineation, thus impairing LV mass assessment. The impact of LV wall thickness for this assessment, for instance in dilated cardiomyopathy, could not be evaluated because of an insufficient subgroup. Nevertheless, we assume that the edge sharpness provided by CS rg should help to distinguish the endocardium from the epicardium as compared to the first-generation CS rt cine that provided blurrier borders. Although a high acceleration factor was demonstrated using CS rg , the improvement of CMR tolerance can only be assumed since this parameter was not evaluated. However, we suppose that the lower number of breath-holdings required by CS rg should help dyspneic or claustrophobic patients undergo CMR examinations. This acceleration might be overestimated and must be interpreted cautiously since the two 1.6 × 1.6 mm 2 in-plane resolution CS sequences were compared with a 1.3 × 1.3 mm 2 in-plane resolution reference sequence. However, Miller et al. demonstrated that the maximal accuracy for functional parameters quantification using SSFP sequence was reached between 1 and 2-mm in-plane spatial resolution [29]. Even though further acquisitions could be performed during CS rg reconstruction during this study (SSFP ref images were acquired first and available), up to 2 min were necessary for images to be reconstructed and displayed to set the next sequences orientation in case of exclusive use of CS rg . Wall motion abnormalities were not evaluated, but the increased temporal resolution provided by segmented acquisition and the absence of interpolation should not impair their visualization on CS rg since the first generation of compressed sensing cine was demonstrated to be reliable for this evaluation (Video S3 (Supplementary Materials)) [44]. The heart rate of the evaluated patients ranged from 54 to 101 beats per minute (R-R intervals: 594 ms to 1111 ms). Since CS rg is a 2-shot sequence and its temporal resolution is constant (37 ms), the amount of data acquired during the acquisition must vary and may have an impact on the quality of the reconstructed cine images and consequently on functional assessment. The impact of heart rate on image quality was not evaluated in this study due to the limited size of the study population. Finally, only sinus rhythm patients were enrolled in this study. Even though a real-time acquisition is more robust than a segmented acquisition versus arrhythmia, the need of only two heart beats per slice should be more arrhythmia-proof than the conventional 8-heart-beat SSFP ref .
Further comparison with other acceleration techniques such as generalized autocalibrating partial parallel acquisition would be interesting since the later requires more heart beats for identical temporal and spatial resolutions [45].

Conclusions
CS rg allows reliable quantification of LV and RV volumes, EF and mass providing similar objective and subjective image quality to SSFP ref . Performed in clinical conditions, CS rg is promising in terms of workflow improvement and image quality recovering in comparison with the first-generation real-time CS rt .
Supplementary Materials: The following are available online at https://www.mdpi.com/article/ 10.3390/jcm10112417/s1, Video S1: Short-axis cine stacks in a 45-year-old woman referred for myocarditis follow-up; Video S2: Four-chamber cine slice in the same patient as Video S1; Video S3, short-axis mid-cavity cine slice in a 68-year-old patient referred for viability assessment after myocardial infarction, demonstrating inferolateral akinesia.
Author Contributions: B.L.: study conception and design, data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content; C.V.G.: data collection and interpretation, critical revision for important intellectual content; A.C.: study conception, critical revision for important intellectual content; L.G.: data collection, interpretation and analysis, drafting of the manuscript; V.S.: data collection and interpretation, critical revision for important intellectual content; J.P.: data collection and interpretation, critical revision for important intellectual content; A.S.: data collection and interpretation, critical revision for important intellectual content; J.H.: data collection and interpretation, critical revision for important intellectual content; M.S.: study conception and design, critical revision for important intellectual content; C.F.: study conception and design, critical revision for important intellectual content; S.T.: study conception and design, critical revision for important intellectual content; D.M.: study conception, critical revision for important intellectual content; F.P.: study conception and design, data collection, interpretation and analysis, drafting of the manuscript, critical revision for important intellectual content. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.

Institutional Review Board Statement:
The study was approved by the research ethics committee of Lille University Hospital.
Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement:
The data presented in this study are available on reasonable request from the corresponding author, subject to approval by the research ethics committee of Lille University Hospital.